Agilink
早期出货规模和投资人质量都很强,但披露仍过薄,难以有把握支撑 $1B+ 估值
Agilink 在灵巧手创业公司里呈现出少见的早期商业化和投资人追捧,但披露单薄、依赖母公司,且估值已拉到 $1B+,因此更适合继续研究。
封面要素
公司概况
Agilink(临界点 AGILINK)是一家位于上海的灵巧机器人手公司,2026 年 1 月从 Agibot 操作业务中拆分出来。公司销售分层 OmniHand 产品组合,覆盖 OmniHand 2025、OmniHand Pro 2025 和旗舰 OmniHand 3 系列,并提供配套夹爪,面向具身 AI OEM、工业自动化用户、物流场景和科研客户。公开报道称,Agilink 在不到五个月内完成四轮融资,到 2026 年 5 月估值超过 US$1 billion,并已出货超过 8,000 只灵巧手和 10,000 个夹爪。上行逻辑在于:它在人形机器人技术栈中战略重要的部件层,罕见地很早跑出了规模;下行风险则是,几乎所有商业、财务和治理细节仍依赖稀疏且由母公司中介的披露。
- 成立时间
- 2026-01-14
- 创始人
- Xiong Kun
- 创立地点
- Shanghai, China
- 总部
- Shanghai, China
- 产品
- 分层灵巧机器人手和夹爪产品,由 OmniHand 2025、OmniHand Pro 2025 和 OmniHand 3 Ultra-T 领衔,产品定位覆盖科研、工业操作、物流、服务机器人和具身 AI OEM 集成。
- 客户
- 具身 AI OEM、工业制造商、物流运营商、研究实验室和服务机器人集成商;除母公司相关部署外,公开点名的第三方证据仍很少。
- 商业模式
- 以硬件为先,销售灵巧手和夹爪;围绕操作数据集、SDK 工具和模型 / 软件附加收入,仍有新选择权。
- 阶段
- Series A+ private
- 融资情况
- 2026 年 1 月至 5 月期间报告了四轮融资,最终在 2026 年 5 月估值超过 US$1 billion;投资方包括 Hillhouse Venture、BlueRun、Baidu Ventures、Yunfeng Capital、SAIC Financial Holdings、Joyson Electronics、Longcheer、C Capital 等。
执行摘要
主要优势
- 截至 2026 年 5 月,公开报道显示已出货 8,000+ 只灵巧手和 10,000+ 个夹爪;对一家刚独立的末端执行器创业公司,这个规模少见。
- Agilink 借力 Agibot 生态,并拿到高瓴、蓝驰创投、百度风投、云锋基金、上汽金控和产业方等一线中国投资人背书。
- OmniHand 产品阶梯从低成本到旗舰款都有覆盖,并把硬件、SDK、触觉传感和未来数据集 / 模型可选项打包在一起。
主要风险
- 没有公开收入、利润率、董事会、股权结构或客户集中度披露,无法用基本面支撑独角兽估值。
- 客户证明仍主要经由母公司 Agibot 的部署体现,外部第三方对 OmniHand 独立需求的证明有限。
- 出口管制、采购禁令和行业泡沫风险可能比当前投资人预期更快压缩市场准入和估值。
未决问题
- 按 OmniHand 和夹爪产品族拆分的经核实收入、毛利率、ASP 和复购数据。
- 客户名单、母公司与第三方客户占比,以及 Agibot 内部项目和外部买家之间的出货量集中度。
- 股权结构表、董事会构成、投资人权利,以及 Agilink 与 Agibot 之间的转让定价或 IP 共享安排。
- OmniHand 3 以及生产级部署的独立可靠性、耐久性和安全验证,且证据不能只来自与龙旗相关的材料。
目录
01公司概览
1.1 身份、拆分背景与商业模式
Agilink(临界点 AGILINK)是 2026 年 1 月从 Agibot 操作业务中拆出的独立灵巧手公司。多家中文媒体引用的工商登记信息显示,上海临界点创新智能科技有限公司于 2026-01-14 在上海成立,法定代表人为邓泰华,Agibot 成立时持股 80%。媒体和 Agibot 官方材料的公开定位一致:Agilink 聚焦灵巧末端执行器,而不是完整 humanoid robots,向具身 AI、工业自动化、物流、服务机器人和科研流程销售五指手、高端操作模块和配套夹爪。创始人兼 CEO 熊坤对多家媒体表示,拆出手部业务,是为了支持持续 R&D、打开 Agibot 生态之外的外部销售,并加快工程驱动的规模化生产。该商业模式意味着收入先来自硬件,之后才是围绕灵巧操作模型和数据集的软件、数据工具机会;不过,收入拆分、定价架构或经常性服务披露均未发布。[CO001, CO002, CO003, CO004, CO005, CO006]
| 指标 | 数值 / 状态 | 日期 | 置信度 | 缺口 |
|---|---|---|---|---|
| 法定设立日期 | 2026-01-14 | 2026-01-14 | 中 | |
| 总部 | 中国上海 | 2026-06-16 | 中 | |
| 母公司 / 来源 | 从 Agibot 灵巧手业务拆分 | 2026-01 to 2026-05 | 中 | |
| 当前阶段 | 私人拆分公司 / 类 Series A+ 私募融资 | 2026-05 | 低 | 未披露正式轮次名称 |
| 最新披露估值 | > USD 1B | 2026-05-19 | 中 | |
| 已披露融资节奏 | 设立后 4 轮 | 2026-05-19 | 中 | 未披露累计融资总额 |
| 灵巧手出货量 | 8,000+ | 2026-05-19 | 中 | |
| 夹爪出货量 | 10,000+ | 2026-05-19 | 中 | |
| 收入 / ARR | 未披露 | 低 | 无公开财务报表或投资人材料 | |
| 员工人数 | 未披露 | 低 | 无公开员工披露 |
公开来源能确认设立、估值门槛、融资节奏和出货数量,但无法确认收入、利润率或员工人数。
[CO001, CO004, CO006, CO024, CO028, CO029]Agilink 夹在 Agibot 母公司资源、灵巧手产品和下游工业用户之间。
[CO005, CO006, CO008, CO023, CO035]1.2 领导层、治理与关键人依赖
对一家刚独立的机器人硬件公司而言,Agilink 公开呈现的领导层格外集中。Agilink 的运营门面是熊坤;他从 Agibot 牵头拆分,并在 2026 年 2 月媒体采访中解释产品层级、商业化重点和独立理由。同一批采访还提到,他曾任职于 Tencent Robotics X、IDEA Research Institute 和 Inovance,并于 2024 年 11 月加入 Agibot,分拆前负责灵巧手业务。但法律控制权似乎仍系在母公司生态上:中国媒体引用的公开工商摘要将 Agibot 董事长兼 CEO 邓泰华列为法定代表人,且 Agibot 在公司成立时仍是控股股东。两者放在一起看,Agilink 在融资和产品路线图上有运营自主权,但启动时尚不具备完全独立的治理画像。未有已审阅来源披露独立董事会、外部董事、创始人股权拆分、清算优先权条款或继任安排。灵巧手技术复杂,熊坤又是对外沟通核心,关键人风险较为实质。[CO011, CO012, CO013, CO014, CO015, CO016]
| 人物 / 实体 | 角色 | 背景或控制权 | 重要性 | 披露缺口 |
|---|---|---|---|---|
| Xiong Kun | 创始人 / CEO / 运营负责人 | 曾任 Tencent Robotics X、IDEA 和 Inovance 高管;2024 年 11 月加入 Agibot 负责灵巧手业务 | 产品路线图和规模化的主要技术商业发言人 | 未公开股权持股或董事会席位 |
| Deng Taihua | 法定代表人;Agibot 董事长兼 CEO | 设立时登记为 Agilink 实体法定代表人 | 表明早期拆分阶段母公司仍保持控制 | 在 Agilink 内部的运营角色未完全说明 |
| Agibot | 设立时控股股东 | 据报持股 80%,出资 400 万元人民币 | 提供供应链接入和客户渠道支持 | 无集团内治理或转让定价细节 |
尽管融资和运营以独立公司形象对外营销,公开记录中的治理仍与母公司相连。
[CO002, CO005, CO006, CO011, CO012, CO013]公开记录最强的是估值、出货和融资节奏,而不是财务报表。
[CO024, CO028, CO029, CO030]1.3 融资历史、所有权与独角兽估值跃升
Agilink 成立后融资节奏异常快。Pedaily 于 2026-01-26 报道,公司已完成两轮融资,由 Hillhouse Venture 和 BlueRun Ventures 领投,并计划追加一轮战略融资。2026-02-10,多家媒体报道新一轮数亿元融资由一家大型互联网公司领投,Baidu Ventures、Yunfeng Capital、Synstellation Capital、Joyson Electronics、Longcheer Technology、SAIC Financial Holdings 等战略与财务投资人参投;老股东 BlueRun 和 Hillhouse 据称加码。到 2026-05-19 和 2026-05-20,Reuters-equivalent 中文商业报道、行业媒体和英文摘要对下一次估值跃升给出趋同说法:Agilink 已完成又一轮数亿元融资,估值超过 US$1 billion,成立不到五个月即成为独角兽。公开记录没有给出的信息同样重要。未有已审阅来源披露累计融资总额、每股价格、董事会权利、员工期权池、债务融资或任何老股转让。因此,融资故事在动能和投资人质量上很强,但在股权结构细节和承保基本面上仍然很薄。[CO006, CO019, CO020, CO021, CO022, CO023]
| 日期 | 事件 | 领投 / 参与方 | 披露规模 | 含义 |
|---|---|---|---|---|
| 2026-01-26 | 宣布两轮初始融资 | Hillhouse Venture、BlueRun Ventures | 未披露 | 验证拆分后即刻存在外部需求 |
| 2026-01-26 | 计划战略融资 | 当时尚未点名 | 未披露 | 表明引入产业资本的意图 |
| 2026-02-10 | 宣布新一轮融资 | 头部互联网公司、Baidu Ventures、Yunfeng Capital | 数亿元人民币 | 标志市场信心上台阶 |
| 2026-02-10 | 产业资本参与 | 参投方:Synstellation Capital、Joyson Electronics、Longcheer Technology、SAIC Financial Holdings | 同一轮的一部分 | 增加制造和下游生态连接 |
| 2026-02-10 | 财务投资人跟投 | C Capital、Wofu、Muhua 等 | 同一轮的一部分 | 将投资人基础拓展到战略企业之外 |
| 2026-02-10 | 老股东支持 | BlueRun Ventures、Hillhouse Capital | 超额认购跟投 | 暗示早期投资人提高信心 |
| 2026-05-19 | 披露最新一轮 | 多家顶级投资人;老股东追加投资 | 数亿元人民币 | 资金指定用于模型、数据集和硬件迭代 |
| 2026-05-19 | 估值里程碑 | 公司层面估值 | > USD 1B | Agilink 五个月内进入独角兽队列 |
公开文章一致描述「数亿元人民币」融资和投资人组合,但未发布精确累计融资总额或 cap table 经济条款。
[CO019, CO020, CO021, CO022, CO023, CO024]1.4 产品栈、出货规模与里程碑
Agilink 的商业化逻辑建立在两个相连主张上:它从 Agibot 手部单元继承了可信产品基础,并且已经把该基础放大到远超实验室体量。公司和媒体材料称,团队在 2025 年量产了多款手部产品,拆分前季度销量已达数千。公开产品引用显示,OmniHand 2025 是一款紧凑型 16-DoF 手,具备 10 个主动和 6 个被动自由度,重量低于 500 克,支持 CANFD 连接和 ROS 2,并有 400 多个触觉感知点;OmniHand Pro 2025 升级到 19 个总 DoF,定位工业操作,价格更高。APC 2026 随后发布 OmniHand 3 Ultra-T,Agibot 将其描述为 22+3 DoF 腱绳驱动旗舰,具备 3D 触觉感知、掌心相机、低于 0.3 秒响应和 10:1 负载自重比。到 2026 年 5 月,Pandaily、Gasgoo、Humanoids Daily 等媒体报道称,累计交付超过 8,000 只灵巧手和 10,000 个夹爪,用例覆盖工业自动化、物流、服务机器人和特种作业。即便如此,公开披露仍缺少 ASP、毛利率、故障率、质保条款和客户集中度,因此出货标题应被视为规模证据,而不是可持续经济性的证明。[CO029, CO030, CO031, CO032, CO033, CO034]
| 日期 | 事件 | 类型 | 参与方 | 含义 |
|---|---|---|---|---|
| 2024-11 | Xiong Kun 加入 Agibot,负责灵巧手业务 | 治理 | Xiong Kun / Agibot | 为后续拆分创造前提 |
| 2025 | 多款手部产品实现量产,季度销量数千 | 规模 | Agibot 内部 Agilink 团队 | 说明业务在正式分离前已经存在 |
| 2025 | 发布 OmniHand 和 OmniHand Pro | 产品 | Agibot 手部业务 | 建立 Agilink 继承的产品家族 |
| 2026-01-14 | 上海临界点创新智能科技有限公司设立 | 创立 | Agilink / Agibot | 独立实体正式起步 |
| 2026-01-26 | 披露两轮融资 | 融资 | Hillhouse Venture / BlueRun Ventures | 早期资本验证拆分论点 |
| 2026-02-10 | 宣布新一轮数亿元融资 | 融资 | 大型互联网领投方、BV、Yunfeng、战略投资人 | 扩大战略股东基础 |
| 2026-04-17 | OmniHand 3 Ultra-T 在 APC 2026 亮相 | 产品 | Agibot / Agilink 关联手部团队 | 提高产品线性能上限 |
| 2026-05-19 | 据报估值超过 US$1B | 融资 | 多家投资人 | 不到五个月达到独角兽状态 |
| 2026-05-19 | 累计交付披露为 8,000 多只灵巧手和 10,000 多个夹爪 | 规模 | Agilink | 商业规模声明公开 |
| 2026-05 | 资金指定用于灵巧手模型和开放数据集 | 产品 | Agilink 管理层 / 投资人 | 表明向硬件加软件护城河移动 |
该时间线合并公司、投资人和媒体里程碑,是本章的权威记录时间轴。
[CO001, CO011, CO019, CO021, CO024, CO028]公开记录显示,Agilink 从内部部门到获得融资的独角兽硬件拆分公司,走得很快。
[CO011, CO019, CO024, CO028, CO029, CO030]1.5 展示材料
02市场分析
2.1 市场边界、范围与替代品
Agilink 的核心市场是灵巧机器人手和智能末端执行器细分:多指、传感器丰富的操作模块,面向 humanoid platforms、移动操作臂和可替换 cobot 附件。市场边界包括硬件(手或末端执行器单元)、嵌入式触觉 / 力感知、专有控制软件,以及支撑 in-context learning 的配套数据采集管线。Agilink 的授权模式把边界进一步扩展到第三方 humanoid integrators 嵌入 OmniHand 平台后支付的 royalties。 Agilink 当前市场不包括:没有嵌入式 AI 或触觉感知的传统单轴或双爪气动 / 电动夹爪(Schunk、SMC、Festo 所覆盖的工业自动化夹爪市场);为刚性治具上重复任务设计的固定臂工业机器人末端执行器;操作需求不需要多指灵巧度的巡检机器人、清洁机器人和服务机器人;以及人工检验或装配服务。 灵巧末端执行器的主要需求拉动来自 humanoid robot platforms 的快速扩张。由于没有独立分析师报告把灵巧机器人手子市场作为离散类别覆盖,市场规模只能借更广义的 humanoid hardware market 作为代理,并假设末端执行器占总平台成本约 10–15%(该估计未被独立验证)。这种结构性依赖意味着,Agilink 的收入机会绑定在 humanoid robots 在制造、物流和服务垂直场景中的采用速度上。 Agilink 替代的主要现状方案是:(1)cobot 机械臂上的标准工业夹爪,需要按任务重新配置;(2)传统自动化无法经济覆盖的操作环节中的人工手部劳动;(3)3C 电子装配中按每个产品 SKU 重新定制的末端执行器。相邻竞争动态包括 soft-robotics 夹爪公司(Soft Robotics、Festo Bionic)、加入自有末端执行器的 cobot 厂商(Universal Robots、Franka Robotics),以及在 Agilink 从 Agibot 拆分后已开始开发的其他中国灵巧手进入者。 更广义 humanoid 市场的分析师估计差异显著,这反映的是真实范围分歧,而不是简单预测错误。Goldman Sachs 口径是 “humanoid robot hardware revenue”,MarketsandMarkets 则覆盖 “humanoid robot market”,包含部分软件价值。IDC 用出货量和 ASP 隐含收入来框定机会。这些方法差异意味着,2025 年 $2.92B–$6B 区间和 2035 年 $15.26B–$38B 区间不能按同口径比较;它们应被理解为对重叠现象的不同观察镜头,而不是单一共识估计。 [CM001, CM002, CM003, CM013, CM038]
| 类别 | 纳入支出 | 排除支出 | 买方 / 付款方 | 与 Agilink 的市场相关性 |
|---|---|---|---|---|
| 灵巧机器人手 | 带触觉 / 力感知、内嵌推理和 SDK 授权的多指 AI 手 | 单轴夹爪、被动挂钩、无 AI 的真空末端执行器 | 人形机器人集成商(直接);制造 OEM(间接) | 核心产品——直接收入和授权来源 |
| 协作机器人操作末端执行器 | 安装在协作机器人手臂上、带力反馈的 AI 多关节末端执行器 | 标准平行夹爪、无内嵌智能的吸盘 | 协作机器人 OEM(Universal Robots、Franka)和工业集成商 | 相邻机会——授权机会;优先级低于人形机器人渠道 |
| 人形机器人平台硬件 | 完整双足人形机器人单元,其中灵巧手是子系统(估计占成本 10–15%) | 轮式 AMR、固定臂协作机器人、四足机器人、无人机硬件 | 制造 OEM(汽车、3C)、物流运营商、服务运营商 | 上游需求拉动——人形机器人采用带动末端执行器需求 |
| 触觉感知和力反馈模块 | 独立触觉传感器阵列、嵌入手部模块的力矩传感器 | 过程传感器、SCADA 仪表、纯视觉感知系统 | 机器人 OEM、研究机构、国防 / 政府实验室 | 相邻机会——可与 Agilink 手部捆绑,也可单独销售 |
| 人形机器人 AI 与具身智能软件 | 面向灵巧操作的专有基础模型;SDK 和数据平台授权 | 通用工业机器人控制软件、PLC/SCADA | 人形机器人集成商和按单台机器人 SaaS 或一次性费用付费的终端制造商 | 通过 Agilink 已宣布的授权模式形成次级收入流 |
市场边界来自 Agilink 产品页面和公开报道。10–15% 末端执行器成本占比是未经验证的内部估计;没有独立市场报告把灵巧手赛道作为独立类别测算规模。纳入 / 排除边界由作者按产品功能和可寻址性界定。
[CM001, CM002, CM003]三层金字塔展示全球人形机器人 TAM、中国灵巧手 SAM 代理值,以及基于当前产品范围和授权模式估算的 Agilink 近期 SOM。
SAM 和 SOM 是作者推导的代理估算;没有独立分析师覆盖单独的灵巧手市场。SAM = 中国人形机器人市场 × 假设末端执行器占比 10–15%(未验证)。SOM 为定性判断;Agilink 未披露收入目标或单位销量。
[CM003, CM007, CM008, CM038]2.2 市场规模——TAM、SAM 与多组分析师口径
公开可访问的独立研究报告中,没有一份把灵巧机器人手或末端执行器子市场作为离散类别覆盖。本章所有测算都借更广义的 humanoid robot hardware market 作为代理,并将 Agilink 可服务部分估算为中国占比和假设末端执行器成本比例的函数。 Goldman Sachs Research(2024)给出最常被引用的机构估计:基准情形下,10–15 年(约到 2035E)humanoid hardware TAM 至少为 $6 billion;中位情形到 2035 年为 $38 billion;蓝天情形最高可达 $154 billion——前提是产品设计、可负担性、技术成熟度和公众接受度等障碍被完全克服。GS 报告将工厂场景的经济可行性校准在 2025–2028 年,将消费场景校准在 2030–2035 年,并要求生产成本每年大约下降 15–20%。ARK Invest 另行估算 humanoid 劳动力替代带来的经济价值为 $24T——这是 GDP 影响口径,不是硬件收入,不能与 GS 数字混同。 MarketsandMarkets(2026 年 4 月 / 6 月)测算全球 humanoid robot market 2025 年为 $2.92B,到 2030 年增至 $15.26B,CAGR 为 39.2%。中国子市场预计为 $0.40B(2025)→ $2.80B(2030),CAGR 为 47.6%,使中国成为增长最快的国家市场。Grand View Research 在付费 URL 上提供了另一组估计;头部数字量级一致,但具体范围不同。 IDC(2026 年 5 月)用单位出货量框定机会:2025 年全球出货 18,000 台 humanoid robots,到 2030 年增长至 510,000+ 台,CAGR 约 95%。按 $28K 单价门槛(Morgan Stanley 买方价格上限)计算,这意味着到 2030 年收入下限约为 $14B,大体与 MarketsandMarkets 估计一致。若按当前约 $100K 的平均售价计算,到 2030 年隐含收入将超过 $51B,高于 Goldman Sachs 中位情形。IDC 明确指出,中国厂商约占 2025 年出货量的 95%。 对 Agilink 的受限 SAM 估计:套用中国子市场份额(2025 年 $0.40B → 2030 年 $2.80B),并假设末端执行器占总平台价值的 10–15%,可得到指示性 SAM:约 $40M–$60M(2025)增长至约 $280M–$420M(2030)——由于缺少独立市场数据,不确定性很高。任何独立分析师报告都未佐证该估计,因此只能把它当作规划代理。 [CM004, CM005, CM006, CM007, CM008, CM009]
| 来源 | 地理范围 | 2025 年估计($B) | 2030 / 2035 预测($B) | CAGR | 指标范围 | 置信度 | 主要限制 |
|---|---|---|---|---|---|---|---|
| Goldman Sachs Research(2024) | 全球 | ~6(基准情景,约 2035) | 2035E:$38B 中性 / $154B 乐观 | 未说明 | 人形机器人硬件收入 | 中 | 长周期预测;2026 年 GS 调研下调了近期出货 |
| MarketsandMarkets(2026 年 4/6 月) | 全球 | 2.92 | 2030 年 15.26 | 39.2% | 人形机器人市场(硬件 + 部分软件) | 中 | 付费报告;只有摘要;范围定义宽于纯硬件 |
| MarketsandMarkets 中国子市场(2026 年 6 月) | 中国 | 0.40 | 2030 年 2.80 | 47.6% | 中国人形机器人市场 | 中 | 同一付费报告限制;中国特定范围 |
| IDC(2026 年 5 月) | 全球 | ~1.5–5(由 18K 单元 × ASP 推算) | 2030 年 >51B(按 $100K ASP);约 $14B(按 $28K 底部 ASP) | ~95%(单位 CAGR) | 人形机器人出货量;收入为推算值 | 高 | ASP 假设带来 3x 区间;收入预测由作者根据 IDC 单位数推导 |
| Grand View Research(2025,付费墙) | 全球 | N/A(付费墙) | N/A | N/A | 人形机器人市场 | 低 | 只能访问摘要条目;未获取完整数据 |
| ARK Invest(经 Forbes/Diamandis) | 全球 | N/A | 2035+ 年 $24T 经济价值(劳动力替代) | N/A | GDP / 劳动力替代价值;不是硬件收入 | 低 | 指标与硬件收入估计根本不同;不应与 GS 相加 |
| Agilink SAM(作者推导代理) | 中国 | 0.04–0.06(代理:中国子市场 × 10–15% 手部成本占比) | 2030 年 0.28–0.42 | N/A | 灵巧手末端执行器子赛道(仅中国) | 低 | 无独立来源;纯代理估计。仅作规划基线。 |
除 Agilink SAM 代理外,所有预测均来自已发布的分析师研究。IDC 收入区间由作者根据 IDC 单位出货预测和两种 ASP 情景推导;IDC 本身没有发布收入数字。ARK Invest 数字衡量经济替代价值,不是硬件收入,不能与 Goldman Sachs 硬件收入估计相加。分析师范围定义并不标准化,因此差异是真实且不可调和的,并非错误。所有数字都应视为三角测算视角,而非共识。
[CM004, CM005, CM006, CM007, CM008, CM009]Goldman Sachs 对 2035 年全球人形机器人硬件收入给出三情景区间,显示不确定性很宽,也说明该口径不能与 ARK Invest 的 $24T 经济替代规模相容。
GS 数字来自 2024 年 Goldman Sachs Research 报告;2026 年 GS 中国调查下调了近期出货预期,但没有修订 2035 年长期估算。ARK Invest 数字仅用于对照;它衡量的是劳动力替代经济价值,不是硬件销售收入,不能与 GS 硬件收入估算相比。
[CM004, CM005, CM006, CM014]2.3 买方、用户与付款方分层
Agilink 处在两层买方结构中。主 go-to-market 渠道里,humanoid robot integrators(Agibot、Unitree、UBTECH、Fourier,以及 Figure AI 或 Apptronik 等潜在未来客户)是 Agilink OmniHand 单元的直接买方;他们把手嵌入自家平台,并承担采购风险。次级渠道里,Agilink 的授权模式可以从围绕 Agilink 软件栈和控制接口自研硬件的 integrators 处获得收入——这是一条独立 B2B 收入流,可降低对单元硬件的依赖。 最积极部署带操作需求 humanoid robots 的终端用户包括:(1)汽车 OEM 制造:UBTECH Walker S2 已确认交付 BYD、Geely、FAW-Volkswagen、Dongfeng Liuzhou Motor 和 Foxconn。预算所有者是工厂工程或制造运营负责人,典型评估周期为 6–18 个月,ROI 锚定在降低劳动力成本。(2)3C 电子制造:Agibot G2 部署于 Longcheer Technology 平板生产设施,据报吞吐量为 310 台 / 小时,成功率 99.9%+;该细分重视精度、速度和快速产品换型。(3)物流与仓储:集装箱搬运、码垛和物品分拣任务,客户要求机器人在中国当前劳动力成本下达到约 50% 人工产能,以实现 2 年回本。(4)服务:酒店、零售导览和餐饮服务——灵巧度要求较低,但部署量大、能带来可见度。 Morgan Stanley AlphaWise 对中国 C-suite 高管的调查(2026)发现,62% 可能在三年内采用 humanoid robots,但只有 23% 对当前产品表示满意,理由是灵巧度、功能和价格存在缺口。关键价格上限是 200,000 元人民币(约 $28K USD),低于该水平时,92% 受访者认为广泛采用可行。这给制造商制造了明确的单位经济性倒逼:要达到 $28K 价格点,既需要硬件降本,也需要软件效率提升,才能在规模化时跑通经济性。 预算所有权呈现一致模式:humanoid 单元的资本开支由制造工程或自动化运营团队控制,ROI 要求来自 CFO 或工厂运营领导层。试点项目通常先由创新或数字化转型预算赞助,之后才进入 capex 线路,形成 12–24 个月的试点到量产缺口,限制近期收入确认。Agilink 获 Crunchbase 确认的 $1B 估值,以及自 2026 年 1 月拆分以来的四轮融资,使其有能力为这段 runway 供血,但试点转化仍是关键收入风险。 [CM022, CM023, CM024, CM025, CM026, CM027]
| 细分 | 终端用户 / 运营方 | Agilink 直接买方 | 付款方 / 预算所有者 | 关键采用触发因素 | 已确认部署 | 价格敏感度 |
|---|---|---|---|---|---|---|
| 汽车 OEM 制造 | BYD、Geely、FAW-VW、Foxconn 工厂运营方 | 人形机器人集成商(UBTECH、Agibot) | 工厂制造工程 / CFO(资本开支) | 劳动力短缺;电动车生产爬坡;自动化税收激励 | UBTECH Walker S2 → BYD、Geely、FAW-VW(2025 年部署) | 中–高;关注在 50% 人类生产率下 2 年 ROI |
| 3C 电子制造 | Longcheer Technology、Foxconn、Samsung 显示装配 | 人形机器人集成商(主要为 Agibot) | 运营 / 自动化负责人 | 产品周期短;手工返工成本;精度要求 | Agibot G2 在 Longcheer 部署(2025–2026);310 单元 / 小时,99.9% 成功率 | 高;单位经济模型必须在当前人民币工资水平下优于人工 |
| 物流与仓储 | 客户:JD.com、SF Express、Cainiao、DHL China | 人形机器人集成商;臂载系统可能直接采购 Agilink | 运营 / 供应链领导层 | 劳动力可得性;旺季吞吐;电商增长 | 已报道早期试点;尚无确认的 Agilink 大规模部署 | 很高;在 50% 人类吞吐下回本期必须达到约 2 年 |
| 服务(酒店 / 零售 / 餐饮) | McDonald's China、酒店连锁、零售品牌 | 人形机器人集成商(UBTECH、Kepler、AGIBOT A2) | 营销 / 运营 | 品牌差异化;一线城市人工成本;顾客新鲜感 | McDonald's 上海机器人服务员(2025–2026) | 低–中等;高端品牌愿意为宣传价值吸收成本 |
| 人形机器人集成商(B2B OEM) | Agibot、Unitree、UBTECH、Fourier、国际厂商(Figure AI、Apptronik) | 直接——Agilink 的主要客户渠道 | 产品工程 / 采购 | 平台 BOM 优化;专用手部能力超过自研水平 | Agilink 嵌入 Agibot 平台(母公司关系) | 中等;集成商把成本传导给终端客户;授权模式降低单机价格敏感度 |
| 政府及国资关联采购 | State Grid、国防邻近设施、智慧城市项目 | 具备政府关系的人形机器人集成商 | 政府采购预算(多年期合同) | 国家安全;基础设施巡检;十五五规划要求 | State Grid 具身智能采购 CNY 68B(规模未验证) | 低;政策驱动;价格相对战略匹配度居次 |
已确认部署来自关于 Agibot(G2/Longcheer)和 UBTECH(Walker S2)部署的公开报道;按细分市场独立验证的 Agilink 品牌手部单元数量尚不可得。State Grid 采购数字(CNY 68B)被多家来源引用,但未能在一手文件中独立核验。 Agilink 在物流场景的具体部署仍仅停留在未确认的试点阶段。基于 Crunchbase 对 Agilink 授权模式的报道,报告假设直接面向集成商是主要渠道。
[CM022, CM023, CM024, CM025, CM026, CM027]双轴矩阵按灵巧度需求(低到高)和近期部署准备度(低到高)定位买家分群,识别 Agilink 最容易切入的象限。
[CM025, CM026, CM027, CM029]2.4 增长驱动因素与政策顺风
五个结构性驱动因素支撑 Agilink 所在市场的近中期增长机会。 第一,中国“十五五”规划(2026–2030)明确把机器人放在现代产业体系核心,并将机器人划入国家战略优先级最高层级。IFR 于 2026 年 5 月确认,下级行业和区域规划必须与该框架对齐,这意味着到 2030 年政府采购和补贴支持将持续。中国已经拥有全球最大的工业机器人装机基数——约 200 万台,是日本的 4.5 倍,并占全球年度工业机器人安装量的 54%。 第二,中国工业机器人国产供应商份额从 2020 年的 30% 升至 2024 年的 57%,验证了结构性进口替代趋势;政府正主动把该趋势延伸到 humanoid robots。这创造了一个受保护的国内需求池,Agilink 这样的中国末端执行器供应商拥有优先进入权。 第三,灵巧操作瓶颈可以说是制约 humanoid robot 商业部署的最大单一技术约束。Agibot 联合创始人彭志辉——同时也是中国 MIIT 人形机器人标准化技术委员会副主任——表示,“人类擅长但传统自动化难以胜任的高价值任务中近 80%”依赖尚未标准化的触觉感知。Agilink 专注这一瓶颈,使其处在 humanoid 生态的关键路径供应商位置。 第四,中国 MIIT 于 2026 年 2 月发布首个人形机器人和具身智能国家标准体系,覆盖六大支柱,包括“肢体及部件”和“安全与伦理”。标准化降低协调成本,推动模块化,并形成稳定接口定义;这比集成式全栈制造商更有利于 purpose-built 部件供应商。 第五,VC 融资浪潮规模空前:全球机器人 startup 在 2025 年融资创纪录达到 $14B,高于 2024 年的 $8.2B;仅截至 2026 年 5 月中旬,中国公司融资已达 $5.6B。这些资本正流向生态公司,包括会成为 Agilink 直接客户的 integrators。Unitree 的 $580M IPO(SSE,2026 年 3 月)及其收入同比增长 335% 至 $235M,验证中国 humanoid 细分已进入商业规模。工厂 cobots 从首个商业版本到批量销售可行用了 7–10 年;humanoid robot ecosystems 迭代更快(每代 6–8 个月,而早期 cobots 以年计),暗示采用曲线会被压缩。 [CM015, CM016, CM017, CM018, CM019, CM020]
| 因素 | 类型 | 强度 | 关键证据 | 时间范围 | 尽调问题 |
|---|---|---|---|---|---|
| 中国十五五规划(2026–2030) | 驱动因素 | 强 | IFR 2026 年 5 月——机器人被置于国家工业体系核心;所有下位规划随之对齐 | 2026–2030 | 确认部委层面面向具身智能 / 灵巧手的具体资金安排 |
| MIIT 人形机器人标准体系(2026 年 2 月) | 驱动因素 | 强 | MIIT 标准覆盖肢体 / 零部件支柱;降低集成摩擦;Agibot 联合创始人进入委员会 | 2026–2028 | 核验已发布标准是否嵌入有利于 Agilink OmniHand 接口规格的要求 |
| 灵巧操作瓶颈 | 驱动因素 | 极强 | 彭志辉(Agibot/MIIT):80% 的高价值任务依赖尚未标准化的触觉感知 | 2026–2028 | 量化人形机器人任务集中,需要灵巧操作而非简单抓取放置的比例 |
| VC 与政策资本流入创纪录 | 驱动因素 | 强 | Crunchbase:2025 年全球机器人融资 $14B;截至 2026 年 5 月中,中国 176 笔交易融资 $5.6B | 2025–2027 | 确认 VC 资金流向零部件 / 末端执行器供应商,而非全栈集成商的占比 |
| 大规模采用的 $28K 成本上限 | 约束 | 极强 | Morgan Stanley 2026 年调查:92% 的 C-suite 要求价格低于 200,000 元($28K)才会广泛采用 | 2026–2028 | 跟踪 Agilink OmniHand ASP 走势和 BOM 降本路线图,与 $28K 整机门槛对比 |
| 仿真到现实能力缺口 | 约束 | 强 | Goldman Sachs 与 IDC 2026:AI 准确率从仿真中的 80–90% 降至真实环境中的低于 50% | 2026–2029 | 索取 Agilink 在 Longcheer 部署和可比产线中的现场成功率数据 |
| NDRC 整合压力(150+ 家竞争者) | 约束 | 强 | NDRC 2026 年警示“高度重复产品”;融资环境转向筛选 | 2026–2028 | 评估 Agilink 相对国内手部厂商的差异化;确认 Agilink 是否在 NDRC 优先名单上 |
| 地缘政治导致市场排除(ASRA 法案) | 约束 | 中等 | 美国参议员在 2026 年 3 月提出 American Security Robotics Act;Agibot 被明确点名 | 2026+ | 跟踪 ASRA 立法进展;评估 Agilink 对国际收入的敞口,以及仅面向中国的上市路径 |
强度评级是基于所引证据的定性判断,并非量化分数。NDRC 警示来源是发言人在新闻发布会上的表态,并非正式监管命令。 截至 2026 年 6 月,ASRA 法案仍是拟议立法,尚未生效。所有时间范围均依据已发布的分析师和政府预测作指示性判断。
[CM015, CM016, CM019, CM030, CM032, CM033]时间线梳理人形机器人和灵巧末端执行器赛道的定调事件,从早期协作机器人类比一直到预计的工业大规模采用。
[CM035, CM039, CM040, CM041, CM031]2.5 采用约束、缺口与负面信号
多个结构性约束压低增长前景,也构成 Agilink 市场渗透的实质风险。 竞争密度和整合风险:中国 NDRC 于 2026 年明确警告“高度重复产品”,并指出中国现有 150 多家 humanoid robot 公司。政府正从普惠补贴转向选择性支持,类似 2018–2022 年 EV 行业整合。Agilink 作为 2026 年 1 月新拆出的公司,必须在预期整合把可存活生态玩家从 150+ 压缩到少数几家之前,建立差异化位置。 能力不成熟和试点到量产缺口:Morgan Stanley 调查数据(2026)显示,只有 23% 的中国 C-suite 受访者满意当前 humanoid robot 产品,主要短板是灵巧度和功能。sim-to-real 缺口仍未解决:Goldman Sachs 中国调查数据和 IDC 均报告,机器人 AI 准确率从仿真中的 80–90% 降至真实场景中的 50% 以下——该失败率使工业部署在精密制造中不可靠。灵巧手需要机器人任何子系统中最高保真的触觉和力反馈,因此最暴露于能力缺口。 地缘政治市场排除:参议员 Cotton 和 Schumer 于 2026 年 3 月提出 American Security Robotics Act(ASRA),众议员 Stefanik 也提出配套众议院法案,拟禁止美国联邦政府采购中国制造的 humanoid robots。Agibot 和 Unitree 被明确点名。即便法案未通过,它也会对美国企业客户形成采购寒蝉效应——类似 DJI 企业无人机和 Huawei 电信先例——可能实际关闭美国联邦政府市场,并减少国防相邻和关键基础设施制造领域的采用。 IFR 技术怀疑:International Federation of Robotics 于 2026 年 5 月判断,“作为通用 humanoid 工厂助手或进入私人家庭的大规模采用,在近中期不会发生。”传统工业机器人在重复性精密任务中更快、更简单、更可靠,仍会是高速制造的骨干。Goldman Sachs 2026 年 1 月中国调查确认了这一方向性转向:行业已从“通用想象”转向安防巡逻、酒店导览和简单物流分拣等窄用例中的“专用落地”。商业规模的通用灵巧操作仍需 1–3 年。 成本上限约束:Morgan Stanley 调查中 92% 受访者提到的 $28K 单机成本门槛(200,000 元人民币),目前低于多数商用 humanoid platforms 的实际售价,更不用说 Agilink 手部组件。达到该价格点需要持续每年 15–20% 的降本——Goldman Sachs 已建模这一轨迹,但其取决于供应链成熟和规模效应,尚未得到保证。 [CM032, CM033, CM034, CM035, CM036, CM037]
2.6 展示材料
03竞争对手
3.1 竞争格局与分类
Agilink 面对的竞争地形,比单纯用灵巧手视角看到的替代方案更宽。买方——humanoid OEM integrators、研究实验室、物流运营商和企业自动化团队——可以用五种方式解决同一操作任务:(1)高端西方 research-grade hand(Shadow Robot Dexterous Hand、Allegro Hand V4),优先考虑已发布规格和生态引用;(2)学术开源平台(LEAP Hand、CMU 2023),用商业打磨换取接近零的采购成本;(3)中国商用独立手(RobotEra XHAND1、Inspire Robots RH56 系列),靠 direct-drive 耐久性或价格竞争;(4)手与平台不可分割的 humanoid OEM 捆绑(Fourier GR-2、带可替换末端执行器的 Unitree R1);或(5)物流末端执行器系统(RightHand RightPick 3、Dexterity AI Mech),把硬件和软件打包给 piece-picking 等单一垂直任务,完全绕开通用手市场。Agilink 位于类别(3)和(4)的交叉点:它从 Agibot 的 humanoid hardware 中拆出,但把手作为独立产品销售给第三方 OEM 和研究人员。今天在中国市场,这一位置独特——Inspire Robots 主要卖向国内工业渠道,面向国际的文档有限;RobotEra 的 XHAND1 则深度绑定 STAR1 humanoid 生态。不过,该位置也让 Agilink 从上下两侧都暴露于替代:上方是资金充足的 incumbents(Shadow Robot 在西方企业买方中的品牌资产),下方是超低价进入者(LEAP Hand 或 Unitree R1 捆绑系统,总价低于 Agilink 独立手标价)。[CP001, CP002, CP003, CP004, CP005]
| 竞争者 | 类别 | 规模 / 融资 | 目标细分市场 | 差异化 | 关键限制 |
|---|---|---|---|---|---|
| Shadow Robot | 西方研究领域老牌厂商 | 1987 年成立;自举 / 机构资金;Humanoid Index 评分 8/10 | 学术研究实验室、高端国防研发 | 24 DOF;品牌资产;事实上的研究引用标准 | 价格很高(>$100K 系统);没有软件部署栈;英国供应链 |
| Allegro Hand V4(Wonik Robotics 产品) | 西方商业研究产品 | 韩国;单价 $24,500;未披露 VC | 学术操作研究;研究生实验室采购 | 16 DOF;成熟 ROS 支持;DC 扭矩控制 | 无触觉感知;每只手约 $24.5K;商业部署证据有限 |
| LEAP Hand (CMU) | 学术开源 | 非商业;MIT 许可证;RSS 2023 论文 | 机器人 ML 研究者;仿真到现实实验室 | $2,000;4 小时组装;基准测试优于 Allegro;开放 SDK(Python、C++、ROS/ROS2、500 Hz) | 无商业支持;无保修;未达到生产级 |
| RobotEra XHAND1 | 中国商业独立产品 | 已融资 $200M+(2026 年 4 月);SF Group、Sequoia China、IDG Capital | 物流 OEM;人形机器人平台集成商;企业自动化 | 12 DOF 直驱;经 Boston Dynamics、NVIDIA、Apple 验证;物流中心作业 1M+ 次 | 12 DOF 低于旗舰同业;与 RobotEra STAR1 平台深度集成 |
| Inspire Robots RH56 | 中国工业零部件 OEM | 2016 年成立;未披露;聚焦中国国内 | 中国人形机器人 OEM;工业自动化零部件买家 | 微型伺服精度;多种 DOF 配置;国内供应链可靠 | 缺少英文文档;国际证据有限;仅 OEM 分销 |
| Fourier GR-2 / FFTAI | 捆绑手部的人形机器人 OEM | 已融资 $200M+;SoftBank 支持;估值 $800M+;员工 800+ 人 | 医疗康复;工业制造;学术 | 每手 12 DOF;6 阵列触觉传感器;总计 53 DOF;兼容 ROS | 整机价格 $100K+;手部不单独销售;上市路径先打医疗 |
| Unitree R1 | 平台捆绑手部 | 已递交 IPO(¥4.2B);2025 年收入增长 335.4%;全球化 | 开发者社区;轻制造;消费级 / 专业消费级机器人 | 起售价 $3,930;2/3/5 指可换末端执行器;15–31 DOF 可配置手臂 | 末端执行器不是独立产品;单手配置 DOF 有限;企业场景未验证 |
| RightHand Robotics RightPick 3 | 物流末端执行器系统 | $23M Series B(2018);之后未披露融资 | 电商履约;医药分销;AS/RS 集成商 | 1,200 picks/hr;99%+ 准确率;已在日本 PALTAC 验证(500K sqft) | 任务特定(非拟人);无触觉;仅适用于单件拣选流程 |
| Dexterity AI Mech | Physical-AI 超人形系统 | 未披露(Fortune 50 客户);与 Beckhoff、NVIDIA、ASRock Rack 合作 | Fortune 50 企业物流;卡车装载;码垛 | 100M+ 次自主决策;Foresight 世界模型;安全评级;TensorRT 提速 17× | 软件系统路线替代独立手部市场;以美国为中心;定价未披露 |
定价来自 humanoid.guide(Allegro、XHAND1、OmniHand)和产品页(Unitree);Shadow Robot 与 Dexterity AI 未公开披露价格。规模 / 融资来自 humanoidindex.org、cntechpost.com 和 roboticstomorrow.com。标记为“未披露”的单元格是真实数据缺口,不代表数值为零。
[CP001, CP002, CP003, CP004, CP006, CP007]九家竞争者按自由度数量(x 轴,0–30 刻度)和商业成熟度得分(y 轴,0–10 序数:1=学术原型,5=有限商业部署,10=量产规模)定位。Agilink OmniHand O10 和 Unitree R1 聚在高成熟度、中等 DOF 象限,Shadow Robot 则以高 DOF 领先但商业规模较低。
商业成熟度是序数,不是度量值;分析师根据公开部署数据、客户证据和融资信号打分。DOF 数量来自官方和第三方规格来源;LEAP Hand 的 DOF 根据公开装配说明估算,因为论文没有给出单一标准值。x 轴封顶在 30,便于阅读。
[CP043, CP044]3.2 西方 research-grade incumbents 与开源挑战者
Shadow Robot Company(伦敦,1987 年成立)在 research-grade 灵巧手中拥有主导品牌;其 Shadow Dexterous Hand 提供 24 个自由度——截至 2026 年 6 月,这是公开记录中 DOF 最高的商业产品——并在 Humanoid Index 上获得 8/10 readiness rating。Shadow Robot 是机器人操作论文中的事实引用标准;顶尖大学实验室采购经理通常会把任何新手与 Shadow 对标。这个品牌护城河很耐久,因为它会自我强化:每篇引用 Shadow 的论文都会延续规格词汇和隐含采购先例。不过,Shadow 系统价格较高,限制其只能进入资源充足的学术实验室和长周期研究项目;在讲究单位经济性的商业生产部署中,它并不竞争。Allegro Hand V4(Wonik Robotics,韩国)提供 16 DOF,标价 $24,500,重量约 1.1 至 1.2 kg,没有触觉或力感知,采用带扭矩控制的 DC 电机。它广泛用于学术操作研究,但面对中国商用替代品时,性价比偏弱;后者如今能以更低成本提供相当或更高 DOF,并带触觉感知。LEAP Hand(CMU,RSS 2023 发表)是研究细分中最具颠覆性的开源竞争者。论文展示它可用现成零件在四小时内组装,成本 $2,000,采用 MIT licensing,提供 Python、C++ 和 ROS/ROS2 API,查询频率最高 500 Hz;在 RSS 2023 论文报告的所有基准实验中,它表现优于 Allegro Hand,而成本仅为后者的八分之一。对于主要标准是低成本数据采集吞吐量的学术买方,LEAP Hand 为任何低于 $15,000 的商业产品提供了可信替代。[CP006, CP007, CP008, CP009, CP010, CP011]
| 能力维度 | Shadow Dexterous Hand | Allegro V4 | LEAP Hand | RobotEra XHAND1 | OmniHand O10 (Agilink) | OmniHand 3 Ultra-T (Agilink) |
|---|---|---|---|---|---|---|
| 主动 DOF | 24 | 16 | 未披露(≥16) | 12 | 16 | 22+3 |
| 重量(kg) | 未披露 | 1.1–1.2 | <0.5(估算) | 1.1 | 0.5 | 0.5 |
| 触觉 / 力感知 | 是(定制) | 否 | 否 | 是(齿轮集成) | 是(400+ 个点) | 是(3D 全手) |
| 掌心相机 | 否 | 否 | 否 | 未披露 | 否 | 是 |
| 商业标价(USD) | 未公开披露 | $24,500 | ~$2,000(零件) | $14,000 | $3,775 | 未披露 |
| SDK / API 开放度 | ROS(专有完整 SDK) | ROS(开源) | Python、C++、ROS/ROS2、MIT 许可 | 未公开记录 | Genie Studio + AimRT(ROS2) | Genie Studio + AimRT(ROS2) |
| 驱动架构 | 气动 / 腱绳 | DC 电机扭矩控制 | DC 伺服 | 全直驱 | 电机 + 齿轮连杆 | 腱绳驱动(22+3 DOF) |
| 独立验证 | 研究论文(多篇) | 研究论文(多篇) | RSS 2023 论文 | Boston Dynamics、NVIDIA、Apple 硬件采用 | Agibot 母公司 10K+ 台 | APC 2026 发布;独立数据有限 |
规格来自 humanoid.guide(Allegro、XHAND1、OmniHand O10/Pro)、arxiv.org(LEAP Hand)、 humanoidindex.org(Shadow Robot)和 agibot.com(OmniHand 3 Ultra-T)。“未披露”单元格反映经直接 URL 抓取确认的真实数据缺口。 价格为标价 / 公开价格;企业 / 批量价格未知。
[CP009, CP010, CP011, CP012, CP015, CP033]3.3 中国商用灵巧手竞争对手
RobotEra(北京,2023 年 8 月成立)是 Agilink 在中国商用细分中最直接的竞争挑战。其 XHAND1 提供 12 个主动 DOF,标价 $14,000,重量 1.1 kg,尺寸 191×94×47 mm,握力能力 25 kg,集成触觉和力感知,并采用齿轮驱动力控关节架构。关键在于,RobotEra 公开把 XHAND1 的全 direct-drive 架构定位为“行业首创”,且在 Agilink 宣布拆分之前,Boston Dynamics、NVIDIA 和 Apple 已采用该硬件——这三家是机器人领域技术要求最高的客户之一。RobotEra 2026 年 4 月合计超过 $200 million 的融资(由物流巨头 SF Group 领投,IDG Capital、Hillhouse、Sequoia China 和 CICC Capital 参投),以及与 China Post 和 SF Express 合作在 10 多个物流中心同时部署机器人,凸显其生态的资金实力和真实落地牵引。RobotEra 于 2026 年 Q2 启动千台级交付,并预计年化增长 300%。Inspire Robots(北京因时机器人科技有限公司,2016 年成立)主要面向中国国内工业市场制造精密微型伺服执行器和灵巧手。产品组合包括 RH56 家族(BFX、DFX、E2、F1)和 EG2 系列夹爪。截至 2026 年 6 月,英文产品文档无法通过直接 URL 抓取访问,限制了独立国际尽调;所有规格和定价信息仅有中文。Inspire Robots 是中国 humanoid integrators 广泛引用的 OEM 部件供应商,意味着它供应由其他公司贴牌的手——这使其既是 Agilink 相邻领域的竞争者,也可能是供应商。Fourier Intelligence(上海,2015 年成立)提供 GR-2 humanoid,总计 53 个自由度,包括每只手 12 DOF 和每只手 6 个阵列式触觉传感器。完整 GR-2 系统价格约 $100,000+,因此不是直接按单手价格竞争的对手;但其 SoftBank 背书(累计融资 $200M+、估值 $800M+、员工 800+、2025 年出货 500+ 台)以及医疗康复和工业制造双重重点,会对企业客户获取形成有意义的渠道冲突。Unitree Robotics(杭州)于 2026 年 4 月推出 R1 双臂 humanoid,起价 $3,930(¥26,900),成为市场上成本最低的完整机器人操作平台。R1 支持可替换末端执行器,包括二指、三指和五指灵巧手,并支持手臂组件 15 至 31 个可配置 DOF。Unitree 报告 2025 年收入增长 335.4%,humanoid 业务成为最大收入板块;公司已向上海 STAR Market 递交最高 ¥4.2 billion 的 IPO 申请。[CP014, CP015, CP016, CP017, CP018, CP019]
| 产品 | 供应商 | 标价(USD) | DOF | 触觉感知 | 包装类型 | 商业可得性 |
|---|---|---|---|---|---|---|
| LEAP Hand | CMU(学术) | ~$2,000(仅零件) | 16+ | None | 开源 BOM + SDK | 研究用途(无商业支持) |
| OmniHand O10 | Agilink / Agibot | $3,775 | 16 | 是(400+ 个点) | 独立手部 + Genie Studio SDK | 量产中(商业) |
| RobotEra XHAND1 | RobotEra | $14,000 | 12 | 是 | 独立手部(针对 STAR1 优化) | 量产中(商业) |
| OmniHand Pro (O12) | Agilink / Agibot | $12,700 | 19 | 是(多轴力) | 独立手部 + Genie Studio SDK | 量产中(商业) |
| Allegro Hand V4 | Wonik Robotics | $24,500 | 16 | None | 独立手部 + ROS 库 | 量产中(商业) |
| Fourier GR-2(系统,捆绑手部) | Fourier / FFTAI | ~$100,000+(整机) | 12(仅手部) | 是(6 阵列式) | 完整人形机器人平台;手部不单独销售 | 量产中(有限) |
| Unitree R1(基础版) | Unitree Robotics | $3,930(系统) | 15–31(手臂);灵巧手选项 | 未披露 | 完整双臂人形机器人平台;末端执行器可更换 | 量产中(商业) |
价格来自 humanoid.guide(OmniHand O10 $3,775;OmniHand Pro $12,700;Allegro $24,500;XHAND1 $14,000)、 unitree.com(R1 基础版 $3,930)、humanoidindex.org(GR-2 约 $100K+)和 arxiv.org(LEAP 约 $2,000)。不含税费与运费。 OmniHand 3 Ultra-T 和 Shadow Robot 价格未公开列示。所有价格均为 2026 年 6 月从官方或权威来源抓取时的价格。
[CP015, CP016, CP022, CP023, CP033, CP034]六家竞争者在八个买家相关维度上的功能覆盖;单元格基于截至 2026 年 6 月获取的来源,显示是 / 否 / 部分 / 未知。
“未知”单元格表示直接抓取来源确认没有数据,不等于假定为负面。LEAP Hand 的 DOF 来自装配文档和性能论文估算。SDK 开放度按许可证类型和公开 API 文档可见性评级。商业保修按产品可获得状态推断;任何竞品都未公开披露保修条款。
[CP011, CP012, CP015, CP035, CP045]3.4 相邻末端执行器系统与现状替代方案
Dexterity AI(美国)代表最成熟的 physical-AI 末端执行器系统替代方案;它竞争的不是手部形态,而是结果。其 “Foresight” world model 已在 Fortune 50 企业客户的生产环境中训练超过 1 亿次自主动作。Dexterity 已在日本 Sagawa Express 的卡车装载作业中部署 Mech superhumanoid,并与 Beckhoff USA 集成 EtherCAT 标准自动化。借助 NVIDIA TensorRT 和定制 CUDA kernels 实现的 17× 推理加速,使 safety-rated physical AI 达到实验室组装手无法匹配的性能水平。Dexterity 的战略位置是 software-and-service-first:它不是靠卖一只手与 Agilink 竞争,而是销售完整操作结果——在任务定义清晰且量大的垂直场景中,实际绕开了独立手市场。RightHand Robotics(2014 年成立,2018 年 12 月完成 $23M Series B)销售 RightPick 3 piece-picking system,最高每小时 1,200 次抓取,具备 Level 4 autonomy 和超过 99% 的抓取准确率,可处理最高 2 kg 物品。RightPick 已部署于 PALTAC Japan(500,000 sq ft,consumer packaged goods)和 Apologistics Netherlands(pharma,AS/RS),并与 UR5e/UR10e cobot arms 集成;它不是拟人手,而是任务专用 picking effector。对物流买方而言,RightPick 3 用更少 DOF 和成熟软件层确定性解决 piece-picking,完全消除了灵巧手需求。多数制造和物流买方的现状替代方案仍是人工劳动加固定自动化(输送系统、真空吸附、刚性双爪夹爪)。灵巧手要跨过的门槛,不只是跑赢某个竞争对手的手;它必须在半结构化环境中跑赢人工灵活性加真空末端执行器的组合经济性,这要求高可靠性,也要求多数运营商的单元成本低于约 $10,000。[CP026, CP027, CP028, CP029, CP030, CP031]
3.5 差异化、切换成本与负面竞争分析
Agilink 的核心性价比优势很具体:OmniHand O10 标价 $3,775(来自 humanoid.guide),而 Allegro Hand V4 为 $24,500——在 DOF 相当或更高(16 vs. 16)且集成 Allegro 完全缺失的触觉感知的情况下,折价 85%。OmniHand Pro(O12)标价 $12,700,具备 19 DOF 和 35 kg 握力,能力曲线位置介于 XHAND1($14,000,12 DOF)和 Shadow Robot 之间。在旗舰层级,OmniHand 3 Ultra-T 具备 22+3 DOF、3D 触觉、掌心相机和低于 0.3 秒响应,是截至 2026 年 6 月中国公开记录中最先进的腱绳驱动手。软件锁定是最粘的护城河。Agilink 的 Genie Studio 开发者平台提供端到端数据采集、一键部署、仿真和数字孪生,以及 9 大类材料、180+ 种产品类型。它支持主流 foundation models(RDT、Pi0、OpenVLA、GR00T),并宣称相对单 GPU 推理带来 2–3× 性能提升。开源中间件 AimRT(v1.7.0,2026 年 4 月)在 Agibot 和 Agilink 部署中提供 ROS2 Jazzy 支持、zenoh、gRPC 和 MQTT。一旦 integrator 在 Genie Studio 上搭建流程,迁移到竞争对手就需要重新训练数据采集管线、重新验证传感器校准,并重写部署代码——即便在早期体量下,这种切换成本也很实质。Agilink 竞争位置的主要负面发现有三点。第一,RobotEra XHAND1 的 direct-drive 架构已被 Boston Dynamics 和 NVIDIA 采用为硬件,这个竞争信号说明 direct-drive 路线已被技术可信客户验证,而 Agilink 尚未公开类似客户引用。第二,Unitree R1 完整系统价格 $3,930 会制造认知问题:想要灵巧操作的买方,可以用大致等于 Agilink 单只手价格的预算买到完整机器人平台。第三,美国立法者于 2026 年 3 月提出法案,拟禁止政府采购中国制造机器人,直接限制 Agilink 进入西方最大企业和政府机器人预算池。该地缘政治逆风同样适用于 RobotEra 和其他中国竞争者,但 Agilink 相比 Fourier Intelligence 的 SoftBank-backed 西方存在,国际销售基础设施更弱,因此影响更重。切换成本和锁定有利于 Agibot 生态 integrators(使用 Genie Studio 和 AimRT),但把 Agilink 集成进非 Agibot 平台的第三方 OEM,相比开源替代品(LEAP Hand、ROS-native Allegro)面临更明显迁移成本。研究买方可以 multi-home,但生产部署中操作复杂,因此更利好 incumbents。[CP033, CP034, CP035, CP036, CP037, CP038]
| 护城河主张 | 竞争威胁 | 严重性 | 缓解措施 / 尽调问题 |
|---|---|---|---|
| 性价比领先(OmniHand O10 $3,775,对比 Allegro $24,500) | Unitree R1 系统 $3,930,削弱独立手部对预算敏感买家的价值主张 | 高 | 核验 Unitree R1 手部选项是否单独定价;跟踪 Unitree SDK 对第三方 OEM 使用的完整度 |
| SDK 与数据平台锁定(Genie Studio、AimRT) | 开源替代方案(LEAP Hand + ROS2)和不依赖 Agibot 的竞争者(XHAND1 已获 Boston Dynamics 验证) 降低感知切换成本 | 中 | 审计 Genie Studio 第三方集成;衡量非 Agibot 机器人本体的 API 迁移摩擦 |
| Agibot 母公司规模(10,000+ 台,APC 2026 伙伴网络) | RobotEra 规模轨迹已经追平:已融资 $200M+、2026 年 Q2 交付 1,000 台,并落地 SF Group 物流部署 | 高 | 按季度比较单元交付节奏;跟踪 Agilink 外部(非 Agibot)客户群是否多元化 |
| 技术差异化(22+3 DOF OmniHand 3 Ultra-T、3D 触觉、掌心相机) | RobotEra 的直驱架构已获顶级科技客户验证;OmniHand 3 与 XHAND1 之间没有已发布的正面对比基准 | 中 | 委托独立操作基准研究;发布 XHAND1 与 OmniHand 正面对比结果 |
| 中国成本结构与制造规模 | 美国立法提案拟禁止政府买家采购中国机器人(2026 年 3 月提出),限制最高价值西方采购细分市场的可触达空间 | 中高 | 跟踪美国 NDAA 和拟议中国机器人采购限制;评估台湾或韩国制造伙伴关系能否满足合规要求 |
严重性评级是分析师基于已抓取来源作出的判断,并非 Agilink 披露的风险。“中高”反映立法结果的不确定性。 尽调问题是建议的下一步核验事项,不是已确认发现。
[CP036, CP037, CP038, CP039, CP040, CP041]围绕企业采购最看重的维度,对比 Agilink 与最接近商业竞品的关键绩效指标。
RobotEra 2026 年合计融资按 10 亿 RMB(以 6.84 汇率折合约 $146M)战略轮加 $200M 新轮估算,约为 $346M。OmniHand 与 Allegro 的价差按 humanoid.guide 标价计算($3,775 对 $24,500)。Agibot 5,100 台出货量来自 humanoidindex.org 引用的 Omdia 2025 数据。
[CP046, CP047, CP048]3.6 展示材料
04财务
4.1 收入架构与定价层级
Agilink 的收入模式以硬件为先:公司向具身 AI 训练、工业自动化、物流、服务机器人和科研流程销售灵巧机器人手、高端操作模块和配套夹爪。公开信息中没有订阅、软件许可或经常性服务产品的定价,因此硬件单位经济性是当前收入估计唯一有意义的杠杆。 来源最精确的价格点是 OmniHand 2025(O10)9,800 元人民币(约 $1,350 USD);Agibot 和独立分销商确认,这是首款价格低于 10,000 元人民币的高自由度灵巧手。第三方经销商 Botonly 截至 2026 年 6 月将 O10 标价 $4,420 USD,显示非中国大陆渠道存在显著加价,但 Agibot 供应链给出的标价仍是 9,800 元人民币。面向工业制造和物流的 OmniHand Pro 2025 仅接受企业询价,因此该层级没有公开 ASP。2026 年 4 月 APC 2026 发布的 OmniHand 3 系列(Ultra-T、Lite、OmniPicker 3)没有确认定价;Lite 被描述为“激进定价”,符合市场份额优先姿态。 Agilink 创始人熊坤把市场分成三段:入门级 6–10 DoF 手,价格数千至 30,000 元人民币;中端 11–17 DoF 手,价格 50,000–100,000+ 元人民币;高端旗舰 20+ DoF 手,价格 200,000+ 元人民币(竞争者包括 Shadow、Tesla 和 Sharpa)。OmniHand 2025 以 9,800 元人民币落在中端细分底部,并大幅低于该区间——这是有意定位,用来拿下科研和首次部署市场。更早的商用灵巧手起价 $10,000(约 73,000 元人民币),最高到 $200,000,意味着 Agilink 入门 ASP 相比此前市场下降 7x–150x。 两条补充收入流在战略层面被披露,但缺少定价或已实现收入数据:向母公司 Agibot 供应手和夹爪,用于 Agibot 品牌机器人集成(内部 / 关联方渠道);以及销售给具身 AI 数据采集流程,由高 DoF 手生成更丰富训练数据。第三条潜在收入流——与 OmniHand 操作数据集绑定的 AI 模型和数据集授权——被熊坤列为 2026 年路线图事项,但尚未成为商业产品。若要确认企业授权收入,需要披露 ASP 和合同细节;这些公开信息目前不存在。[CI001, CI013, CI014, CI015, CI016, CI017]
| 收入流 | 机制 | 单位 | 当前状态 | 收入质量 | 尽调问题 |
|---|---|---|---|---|---|
| 灵巧手硬件销售(O10) | 按单元以标价销售 | 9,800 RMB/unit | 活跃;截至 2026 年 5 月已出货 8,000+ 台 | 中——标价已确认,实际定价组合未知 | 实际 ASP 组合、已开票与已转让单元、退货率 |
| 灵巧手硬件销售(Pro 2025) | 企业按单元报价 | 企业报价;无标价 | 活跃;销量未知;面向工业 / 物流 | 低——价格或销量均未公开 | 确认 ASP、单元销量和客户名称 |
| OmniHand 3 系列硬件销售 | 按单元;发布定价待定 | 未披露(Lite “激进定价”) | 发布阶段;2026 年中开始交付 | 低——无确认 ASP 或订单簿 | OmniHand 3 Ultra-T 标价与积压订单 |
| 夹爪销售(OmniPicker) | 按单元,ASP 低于手部 | 未披露 | 活跃;截至 2026 年 5 月已出货 10,000+ 台 | 低——无 ASP,无产品级拆分 | 夹爪 ASP、毛利率和应用拆分 |
| 向关联方 Agibot 供货 | 按 Agibot 平台成本内部转让 | 转让定价(未披露) | 可能重要;Agibot 持有 Agilink 80% | 低——转让价格可能不等于市场 ASP | 披露关联方收入占比和定价政策 |
| AI 模型与操作数据集授权 | 未来授权或 SaaS;尚未商业化 | 暂不适用 | 仅路线图;Xiong Kun 提到 2026 年 | 不适用——尚无商业产品 | 首笔授权交易预计时间;定价模型 |
所有单元数量和 ASP 均来自截至 2026 年 5–6 月的公开第三方和公司关联来源。收入质量分数是作者对证据强度的判断。 关联方收入占比未单独披露,可能构成出货量的大部分。
[CI013, CI015, CI016, CI018, CI019, CI020]| 产品 | 标价(RMB) | 标价(USD 等值) | 定价依据 | 来源置信度 | 备注 |
|---|---|---|---|---|---|
| OmniHand 2025 (O10) | 9,800 | ~$1,350 | 公开标价 | 中 | 首款低于 10K RMB 的高 DoF 灵巧手;robotscanada.com 和 Agibot 商店页面已确认 |
| OmniHand 2025 (O10) — 第三方经销商 (Botonly) | ~32,200 | $4,420 | 第三方经销商加价 | 中 | 约为标价的 3.3 倍;体现国际市场溢价 |
| OmniHand Pro 2025 | 企业报价 | 企业报价 | 仅面向企业;无公开标价 | 低 | 估计可能为 30,000–100,000 RMB;未确认 |
| OmniHand 3 Ultra-T | 未披露 | 未披露 | 发布时未公布价格(2026 年 4 月) | 低 | 高端旗舰;可能瞄准 50,000–200,000 RMB 区间(作者估计) |
| OmniHand 3 Lite | 未披露 | 未披露 | 被描述为「定价激进」 | 低 | 可能等于或低于 OmniHand 2025 水平;未确认 |
| OmniPicker 3(工业夹爪) | 未披露 | 未披露 | 未公布价格 | 低 | 低于灵巧手;估计 2,000–8,000 RMB |
| 市场层级:入门(6–10 DoF) | 2,000–30,000 | $280–$4,100 | Xiong Kun 市场刻画 | 中 | 展示 / 演示细分;OmniHand 2025 定价 9,800,低于其 DoF 类别的这一区间 |
| 市场层级:中端(11–17 DoF) | 50,000–100,000+ | $6,900–$13,800+ | Xiong Kun 市场刻画 | 中 | 运营部署;此前 16-DoF 手的市场价格为 $10K+ |
| 市场层级:高端(20+ DoF) | 200,000+ | $27,500+ | Xiong Kun 市场刻画 | 中 | 研究 / 精密场景;Shadow、Tesla、Sharpa 在此竞争 |
标价按公开披露口径列示。所有企业价和第三方价格均来自独立确认或估计。USD 等值按约 7.26 的 CNY/USD 汇率计算。市场层级数据来自 aiflashdesk.com 对 Xiong Kun 的引用。
[CI013, CI014, CI015, CI016, CI017, CI018]客户活动通过分层产品漏斗转成硬件收入;到 2026 年中,软件和数据流仍只是路线图项目。
毛利节点为估算;除 O10 的 9,800 RMB 标价外,所有财务值均未披露,基于行业基准推算。
[CI013, CI022, CI025, CI026]4.2 资本形成与投资人结构
2026 年 1 月至 5 月,Agilink 连续完成四轮融资,节奏异常快;SCMP 分析师 Wu Meimei(ITJuzi)称其在仿人机器人部件赛道「前所未有」。前两轮在 2026 年 1 月底前完成,由 Hillhouse Venture 和 BlueRun Ventures(BRV)领投。第三轮于 2026 年 2 月 10–11 日宣布,融资数亿元人民币,领投方是一家大型互联网公司,参投方包括 Baidu Ventures、Yunfeng Capital(Yunfeng Fund LP)、C Capital、Synstellation Capital、Joyson Electronics(Ningbo Joyson Electronic Corp.)、Hangzhou Longqi Scientific Investment(Longcheer Technology)、SAIC Motor Financial Holdings、BRV Partners,以及老股东 Hillhouse 和 BlueRun。C Capital 也在官网单独确认投资,并把 Agilink 定位为具身智能关键部件公司。第四轮于 2026 年 5 月交割,再融资数亿元人民币;披露投资方包括 Baidu Ventures、Yunfeng Capital、SAIC Financial Holdings 和 Hillhouse Capital。 据 SCMP、Reuters、Humanoidsdaily 以及多家中文商业媒体报道,第四轮后公司估值越过 10 亿美元(独角兽门槛)。公司没有公开披露累计融资总额;不过四轮中至少三轮被描述为「数亿元人民币」或更大规模,合计金额合理推测已超过 5 亿元人民币(约 $68 million),且可能高得多。由于收入代理值本身不确定(见第 4 节),估值 / 估算收入倍数高度依赖假设。 需要把 Agilink 的融资与母公司 Agibot 的融资区分开。Agibot 在 2024–2025 年另行融资约 $225 million 等值资金,并正在推进 STAR Market IPO。Agilink 设立时,Agibot 持股约 80%,但之后四轮外部融资带来的稀释幅度没有量化。战略投资者阵容包括 SAIC Financial Holdings、Joyson Electronics、Longcheer Technology,说明汽车和电子供应链深度参与者押注灵巧手会成为下一代产线的标准部件。SAIC 和 Joyson 都是上市制造商,提前锁定优先供应关系有明确战略逻辑。Joyson 自身 2025 年年报(提交至 CSRC 指定信披平台)确认其向「新兴智能体供应链」战略扩张,与投资 Agilink 的姿态一致。[CI002, CI003, CI004, CI005, CI006, CI007]
| 投资人 | 投资人类型 | 轮次 | 战略理由 | 独立性 |
|---|---|---|---|---|
| Hillhouse Capital / Hillhouse Venture | 财务(VC/PE) | 第 1–2 轮 + 第 4 轮(返投) | 中国头部 deep-tech VC;广泛看好机器人 | 财务 |
| BlueRun Ventures (BRV Partners) | 财务(VC) | 第 1–2 轮 + 第 3 轮(返投) | 聚焦中国的早期科技投资人 | 财务 |
| Baidu Ventures | 战略(Corporate VC) | 第 3 轮 + 第 4 轮 | Baidu AI 生态;具身 AI 机器人战略 | 战略 — 与 Baidu AI 平台重叠 |
| Yunfeng Capital(Yunfeng Fund LP 基金) | 财务 / 战略 | 第 3 轮 + 第 4 轮 | Jack Ma 关联基金;科技 portfolio 覆盖广 | 财务 / 战略 |
| C Capital | 财务(VC) | 第 3 轮(已确认) | 2023 年起即为 Agibot 早期投资人;加深生态敞口 | 财务 |
| Synstellation Capital | 财务(VC) | 第 3 轮 | 中国 deep-tech 投资人 | 财务 |
| Joyson Electronics (Ningbo Joyson) | 战略(Corporate) | 第 3 轮 | 汽车电子 / 智能车身供应链;潜在客户 | 战略 — 潜在买方关系 |
| Longcheer Technology(Hangzhou Longqi Scientific Investment 载体) | 战略(Corporate) | 第 3 轮 | 电子制造;可能与 Agilink 零部件供应产生协同 | 战略 |
| SAIC Motor Financial Holdings | 战略(Corporate) | 第 3 轮 + 第 4 轮 | 中国最大车企;关注机器人和自动化供应链 | 战略 — 潜在汽车场景部署客户 |
投资人名单汇总自 MarketScreener(S&P Capital IQ 的第 3 轮数据)、C Capital 确认信息,以及 Humanoidsdaily 对第 4 轮 的报道。按 Pedaily 披露,第 1–2 轮参与方仅限 Hillhouse 和 BlueRun。完整 cap table 未公开。持股比例、领投 / 跟投身份和董事席位均未披露。
[CI004, CI005, CI006, CI007, CI008, CI009]以 8,000+ 台出货和至少三轮数亿元融资为锚,给出 Agilink 累计收入和隐含募资额的保守、基准、高位情景;所有数字均为估算。
所有数字均由作者根据公开出货数据和行业 ASP 基准估算。收入区间使用 CNY/USD 7.26。实际 ASP 组合、关联方转让定价和保修计提未知,可能让这些数字向任一方向显著偏移。
[CI010, CI019, CI022, CI023, CI032, CI037]4.3 单位经济模型与硬件成本结构
Agilink 或 Agibot 均未公开披露任何灵巧手产品的毛利率、销售成本或单台盈利数据。本节所有单位经济模型分析均为估算或推断,并已相应标注。 中国中端灵巧机器人手的物料清单(BOM)大致可以按行业分析师估算拆分:驱动电机占 BOM 成本 30–40%,传感器和电子器件(触觉阵列、PCB、MCU、边缘处理)占 20–30%,结构材料(注塑 PA / 硅胶加铝合金)占 20–25%,线缆和齿轮占 5–10%,装配与人工占 5–15%。若 ASP 为 9,800 元人民币,且毛利率为 30–55%,这些比例意味着单台材料成本约 4,000–7,000 元人民币。30–55% 的毛利率符合硬件初创公司常见水平,但 9,800 元人民币的激进入门定价——公司描述为扩大可及性而非最大化利润——可能压缩毛利。中国灵巧手厂商受益于本土微型电机(规模采购时每个 $10–$30)和注塑件供应链,西方竞争者很难以同等成本获得。 Agilink 的一个关键经济性差异来自自研电机和关键材料部件;Xiong Kun 称这是成本控制和「核心定价权」的基础。垂直整合降低了对第三方执行器供应商的依赖,而执行器往往是 BOM 中最大单项成本。Agilink 还可借助母公司 10,000 台生产里程碑积累的 Agibot 供应链基础设施和制造规模,进一步压低边际成本。上述因素合起来说明,Agilink 维持竞争性定价的时间可能比独立初创公司更长,但精确单台成本和产能数据仍未公开。 营运资本压力不小:硬件制造需要库存融资、部件采购提前期和产线资本开支。初创公司若在约 12 个月内从发布前爬坡到 8,000+ 台,意味着大量库存融资和产线投入,与其快速融资节奏一致。来自 Agibot(关联方买家)的客户预付款会减轻营运资本压力,但规模未知。[CI025, CI026, CI027, CI028, CI039, CI036]
| 指标 | 数值 / 估计 | 置信度 | 重要性 | 尽调要求 |
|---|---|---|---|---|
| OmniHand 2025 标价 ASP(RMB) | 9,800 | 中 | 用收入 proxy 推算单台收入下限 | 确认扣除折扣和关联方调整后的实际 ASP |
| 混合 ASP(估计,手部产品组合) | 15,000–25,000 RMB(估计) | 低 | Pro 占比越高,收入 proxy 会明显抬升 | 要求按 SKU 提供出货量,以估算混合 ASP |
| 硬件毛利率(估计) | 30–55%(估计) | 低 | 决定现金生成能力和 R&D 投入空间 | 要求经审计 P&L 或投资人级管理账 |
| OmniHand 2025 单台 BOM(按 COGS 的 40% 估计) | 约 3,900 RMB(约 $540 USD),估计 | 低 | 理解生产经济性的关键 | 确认单台 BOM 成本和零部件采购条款 |
| 执行器电机成本(估计) | BOM 的 30–40%(单台约 1,200–1,600 RMB) | 低 | BOM 中最大单项;垂直整合电机可能压低该成本 | 披露自研电机成本与市场价格的对比 |
| CAC(获客成本) | 未披露 | 低 — 不适用于硬件初始部署 | 关系到生态渠道与直销渠道的拆分 | 识别主要销售渠道和销售周期 |
| 月度 burn rate(估计) | 15–40 million RMB / 月(估计) | 低 | 决定资金 runway 和下一轮融资时点 | 要求管理层现金流预测 |
| 估计 runway(第 4 轮后) | 12–36 个月(估计) | 低 | 按下限,2027 年前需要下一轮融资 | 确认现金余额和当前 burn rate |
| 营运资本强度 | 高(硬件制造) | 中 | 库存和应收账款带来现金流压力 | 库存周转率和 DSO |
| 关联方收入占总收入比例 | 未知(可能较重要) | 低 | Agibot 内部渠道会扭曲商业定价信号 | 披露来自 Agibot 及其关联方的收入占比 |
所有非 ASP 指标均由作者按中国灵巧手硬件创业公司的行业基准估计。公司未公开披露财务报表或管理账。标注「估计」的数值不确定性很高。
[CI013, CI022, CI023, CI025, CI026, CI027]证据支撑漏斗顶端的 ASP 和中段的战略成本优势;毛利以下没有公开数据,单位经济模型在这里断开。
ASP 以下所有节点均为估算。没有经审计财务。BOM 百分比是 9,800 RMB 价格带中国灵巧手硬件的行业基准区间。
[CI013, CI026, CI027, CI028, CI039]4.4 基于出货量的收入代理值与牵引力缺口
Agilink 公开披露的唯一定量牵引力指标是出货量:据 Pandaily、Gasgoo、Humanoidsdaily 和 Agibot 自有材料,截至约 2026 年 5 月 19 日,已交付 8,000+ 只灵巧手和 10,000+ 只夹爪。同期约 1,000 台交付设备在零售仓、药房和工厂以 8 小时连续班次运行,提供了粗略利用率信号。 可基于这些出货量构造收入代理值。OmniHand 2025(O10)单价为 9,800 元人民币。如果全部 8,000 只手都按这一 ASP 底价售出,累计收入约 7,840 万元人民币(约 $10.7 million)。但真实数字有多重不确定性。第一,产品组合包括 OmniHand Pro 2025(企业定价,ASP 未知,可能为 30,000–80,000 元人民币或更高)。第二,部分灵巧手可能在 Agibot 生态内部生产并转移,采用内部转移价格,而非外部商业 ASP。第三,夹爪(10,000+ 只)ASP 低于灵巧手,可能每只 2,000–5,000 元人民币。若按 50% O10、ASP 9,800 元人民币,30% Pro、ASP 30,000 元人民币,20% 旧款 / 内部转移、ASP 5,000 元人民币测算,平均手部 ASP 约 17,000 元人民币,对应累计灵巧手收入接近 1.36 亿元人民币(约 $18.6 million)。再加上 10,000 只夹爪、每只估算 3,000 元人民币,可新增 3,000 万元人民币。由此得到粗略累计收入区间 5,000 万–2 亿元人民币($7–27 million),且对 Pro 占比假设高度敏感。所有估算都必须附带一个限制:拆分前转移和关联方销售未必代表公平市场收入。 关键问题是,Agilink 没有独立客户收入数据,没有应收账款披露,也没有公开经审计财务报表。出货量本身也只见于公司同向媒体和行业数据库;没有独立审计或第三方验证发布。客户集中度(多少流向 Agibot 自有机器人、多少卖给外部买家)、退货 / 故障率、保修义务和递延收入处理等证据缺口,对投资判断都很重要。[CI019, CI020, CI021, CI022, CI023, CI024]
| 缺失指标 | 重要性 | 可用证据 | 具体尽调路径 |
|---|---|---|---|
| 收入(总额及按 SKU) | 没有收入就无法承销估值倍数 | 只有出货量;无财务报表 | 要求管理账;若已申报,查找 STAR Market IPO 招股书 |
| 按产品线拆分的毛利率 | 硬件毛利质量决定 R&D 资金能力 | 仅有行业基准估计 | 要求经审计 P&L;对比 Unitree 和 UBTECH IPO 时的公开披露 |
| 月度 burn rate | 决定 runway 和对下一轮融资的依赖 | 无披露;按规模基准估计 | 要求投资人级现金流预测 |
| 现金及现金等价物 | 验证 runway 估计 | 未披露 | 要求资产负债表;也可核查是否已向 CSRC 递交 IPO 招股书 |
| 客户集中度与关联方收入 | Agibot 既是大股东,又可能是最大客户,存在冲突 | 仅有定性推断 | 要求关联方交易明细;审阅所有 Agibot-Agilink 商业协议 |
| 第 4 轮后 cap table | 四轮融资带来的稀释不清楚;Agibot 剩余持股未知 | 无公开披露;设立时 Agibot 约持股 80% | 通过国家企业信用信息公示系统审阅工商登记更新 |
| 员工人数与组织结构 | 规模信号和 burn 指标 | 未公开披露 | LinkedIn 员工数调查;直接询问公司 |
| 质保与故障率准备 | 硬件责任影响真实毛利率 | 无披露 | 要求产品质保条款和历史故障 / 退货数据 |
| 与 Agibot 的转让定价 | 内部销售若不按市场价,会扭曲商业牵引力 | 无披露 | 要求所有集团内销售协议和定价基准 |
对一家 IPO 前公司来说,这些指标天然偏私密。上述缺口对中国 Series A–B 阶段硬件创业公司并不罕见,但没有投资人级访问权,就无法承销。
[CI024, CI035, CI036, CI040]将 Agilink 的四类主要资本强度向量映射到现金流时点、规模和可信度,展示硬件优先型分拆公司常见的营运资本和资本开支画像。
所有规模数字均为作者使用中国灵巧手硬件制造商行业基准估算。公司未披露资产负债表或现金流量表。所有标记为“未验证”的行都需要投资人级别披露来确认。
[CI003, CI037, CI038, CI039, CI040]4.5 资本充足性与烧钱节奏
Agilink 密集融资,不到五个月完成四轮。这种节奏即便按中国科技行业标准也不寻常,既反映制造扩产经济性的紧迫性,也反映投资人希望在竞争壁垒关闭前抢占灵巧手品类。此阶段硬件初创公司的烧钱速度通常为每月 $500,000–$3 million,主要由工程团队薪酬、制造资本开支、部件采购和研发驱动。Agilink 计划资金用途(据公司口径来源)包括:开发可靠的通用灵巧执行平台、扩大规模化生产和交付能力、建设全栈 AI 模型 / 数据能力。三条线都吃资本。 Agilink 未披露现金余额、月度烧钱速度、债务融资额度或 runway。按估算累计融资额(基于至少三轮「数亿元」融资,可能 >5 亿元人民币),并假设硬件初创公司月烧钱 1,500 万–4,000 万元人民币(按制造密集型研发放大),第四轮完成时(2026 年 5 月)的 runway 约为 12–36 个月。下限意味着 2027 年底前资金需求会加速到来;上限假设运营精简且预收款回款强。下一轮触发点更可能绑定产线规模、新产品发布里程碑(OmniHand 3 放量)和潜在国际市场进入,而非盈利。公司没有披露债务或项目融资安排。 资本充足性的关键风险在于,收入代理值转化为已收现金的速度。如果 Agibot 生态既是主要股东又是主导客户,来自母公司的贸易应收款可能累积,但外部现金流入并未加速。公开披露没有量化这种关联方动态。Joyson Electronics、SAIC Financial Holdings 和 Longcheer Technology 都是战略投资者,也都代表潜在未来客户;这在一定程度上扩大了潜在直接客户群、缓解集中风险,但 Agilink 与这些投资者之间尚无公开确认的商业协议。[CI003, CI037, CI038, CI039, CI040, CI036]
| 项目 | 数值 / 状态 | 置信度 | 备注 |
|---|---|---|---|
| 总累计融资额(仅 Agilink) | 未公开披露;估计 >500M RMB | 低 | 四轮中至少三轮被描述为「数亿元」 |
| 最新投后估值 | >$1 billion USD(独角兽) | 中 | SCMP、Reuters、Humanoidsdaily 和多家中文媒体报道(2026 年 5 月) |
| 账上现金 | 未披露 | 低 | 未公开披露;考虑到近期融资轮(2026 年 5 月),假设金额较充足 |
| 月度 burn rate(估计) | 15–40 million RMB / 月 | 低 | 按硬件创业公司工程和制造爬坡惯例估计 |
| 估计 runway(第 4 轮后,2026 年 5 月) | 12–36 个月 | 低 | 假设累计融资 500M+ RMB、月 burn 15–40M |
| 计划资金用途 | 通用灵巧执行平台;量产爬坡;AI 模型 / 数据开发 | 中 | 公司口径;见 Humanoidsdaily 对第 4 轮的报道 |
| 下一轮触发条件 | 未披露;可能绑定 OmniHand 3 放量里程碑和潜在 IPO 准备 | 低 | 推测性判断;无公开披露 |
| 债务 / 项目融资 | 未披露 | 低 | 未发现公开债务授信或项目融资披露 |
| 母公司 Agibot 背景(不同于 Agilink) | Agibot 融资约 $225M 等值(2024–2025);正在推进 STAR Market IPO | 中 | Agibot 资本不是 Agilink 资本;此处仅作背景标注 |
所有 burn rate 和 runway 数字均为作者按硬件创业公司行业基准估计。Agibot 母公司背景行已明确标注,并与 Agilink 自身资本状况区分。到 2026 年 6 月,Agilink 尚无经审计财务报表或投资人级披露。
[CI003, CI010, CI012, CI033, CI034, CI037]4.6 财务结论与尽调阻断项
Agilink 的财务画像有吸引力,但高度不透明。收入架构说得通,硬件定价策略也明显为放量牺牲短期毛利,符合先铺装机量、再通过软件、数据和升级周期扩张的 land-and-expand 模型。基于出货量的收入代理值(截至 2026 年 5 月累计 $7–27 million)合理,但完全建立在无法独立验证的假设上,因为公司没有任何经审计财务披露。 即便采用乐观假设,估值 / 估算收入比也意味着超过 40x 远期收入的高度投机倍数。只有一种逻辑能支撑它:Agilink 可能成为全球仿人机器人行业的主导灵巧手供应商,而这个赛道已被中国政府规划者明确瞄准,尽管他们也警示了泡沫风险。NDRC 2025 年警告称,150 多家机器人公司因重复产品面临「速度与泡沫」风险;这一警告直接适用于中国灵巧手细分市场,至少五家资金充足的公司(LingXin、Linkerbot、XHAND、Shadow 等)都在争夺同一批机器人平台集成机会。 财务尽调阻断项很严峻:没有经审计收入,没有毛利率披露,没有烧钱速度,没有股权结构表,没有客户集中度分析,也没有独立董事会。母公司 Agibot 在设立时控制约 80% 股权,引发对转移定价安排、治理独立性和客户合同潜在利益冲突的合理担忧。投资者基础包括战略产业方(SAIC、Joyson、Longcheer),它们与 Agilink 的商业关系会让公平交易分析更加复杂。在 Agilink 提交公开发行申请或向投资者披露财务之前,其财务故事主要仍是市场位置押注,而不是可承保的投资。[CI010, CI011, CI024, CI030, CI031, CI032]
4.7 展示材料
05产品与技术
5.1 OmniHand 产品族与定位
截至 2026 年 6 月,Agilink 仍通过母公司 agibot.com 店铺和域名运营,对外提供 OmniHand 品牌下三档产品,另有配套夹爪线。OmniHand 2025——市场名称为「Agile Edition」——是可及的入门档,面向交互式服务机器人、研究实验室和教育场景。它拥有 16 个总自由度(10 个主动、6 个被动)、重量 ≤500 g、外形尺寸 180×85×38.5 mm,设计上可适配标准仿人机器人腕部接口和协作机器人臂,不需要结构加强。其发布价为 ¥9,800(约 $1,350 USD),被广泛称为首个跌破 ¥10,000 价位的高 DoF 灵巧手,扩大了研究和开发者市场的可及性。Botonly 经销商标价约为 $4,420 USD,反映国际加价。 OmniHand Pro 2025 是工业档,拥有 19 个 DoF(12 个主动、7 个被动)、重量 ≤750 g,并凭借电机 + 丝杆传动实现显著更高的 35 kg 负载能力(掌心向上)。Pro 搭载多模态传感:3 轴指尖力、1 轴掌心力,以及 150+ 个触觉点,灵敏度 0.01 N,比基础款细一个数量级。humanoid.guide 标价 $12,700 USD;它以能力对标价格高得多的研究级灵巧手,同时仍远低于 Shadow Dexterous Hand 等平台的六位数价格。 OmniHand 3 Ultra-T 于 2026 年 4 月 17 日在 Agibot 2026 Partner Conference(APC 2026)发布。公司称其采用腱驱动方案,拥有 22+3 个 DoF,重量仅 500 g——与入门款 OmniHand 2025 相同,尽管主动自由度超过后者两倍。除 Ultra-T 外,公司还发布了 OmniHand 3 Lite(加固高冲击版本)和 OmniPicker 3(工业夹爪,140 N 力、1,000,000 次循环耐久)。Ultra-T 尚未披露商业上市日期;所有已发布规格均来自 Agibot 新闻稿,未被评测方、客户或第三方基准独立验证。[CE001, CE002, CE003, CE009, CE010, CE014]
| 产品 / SKU | 目标用户 | 状态 / 成熟度 | 关键差异化 | 尽调缺口 |
|---|---|---|---|---|
| OmniHand 2025 (O10, Agile) | 服务机器人、研究、教育 | 商用 — 2025 年起量产,标价 ¥9,800 | 首款低于 ¥10,000 的高 DoF 手;16 DoF(10 个主动);CANFD/RS485;可选触觉 | 无可靠性 / MTBF 数据;质保未披露 |
| OmniHand 2025(带触觉) | 研究、精细操作 | 商用 — 外接触觉模块,≤550 g | 400+ 触觉点,0.1 N 分辨率;支持手内滑移检测和安全感知 | 未公布传感器故障率;无 IP 保护时间线 |
| OmniHand Pro 2025 (O12) | 工业自动化、精密装配、物流 | 商用 — 已量产,2025 年 10 月发布会上搭载 G2 展示 | 19 DoF(12 个主动);35 kg 负载;0.02 mm 重复精度;$12,700 USD | 无第三方负载或耐久性验证;企业报价不透明 |
| OmniHand 3 Ultra-T | 高灵巧装配、多轴操作 | 已发布 — APC 2026(2026 年 4 月 17 日);无商用上市日期 | 22+3 腱绳 DoF;500 g;10:1 负载比;掌心相机;3D 触觉(均为公司说法) | 无独立规格验证;无定价;无交付时间线 |
| OmniHand 3 Lite | 高冲击工业环境 | 已发布 — APC 2026;细节稀疏 | 面向耐久优先环境的加固版 | 无规格、无定价、无上市信息 |
| OmniPicker 3(夹爪) | 制造产线、装配 | 已发布 — APC 2026 | 140 N 夹持力;1,000,000 次循环耐久;模块化触觉 | 不是灵巧手;无独立验证 |
状态基于公开商用可得性和展示证据,而非工程就绪度说法。到 2026-06-16,OmniHand 3 Ultra-T、3 Lite 和 OmniPicker 3 仍只是公司发布的产品,尚无确认发货日期。
[CE001, CE009, CE017, CE022, CE027, CE028]三层栈从物理机械,经传感和驱动,到软件与 AI 平台,展示 OmniHand 硬件层如何接入 Agibot 开发生态。
层级边界为分析划分;具体组件来自官方产品页、GitHub SDK 和 Genie Studio 文档。Ultra-T 层级项标记为已发布但未验证。
[CE003, CE006, CE023, CE024, CE025, CE029]5.2 硬件架构——驱动、传感与机械设计
OmniHand 2025 和 Pro 2025 都采用连杆机构传动,每个主动自由度由独立电机驱动。基础款使用电机 + 齿轮;Pro 使用电机 + 丝杆,以 250 g 重量代价换来更高负载能力。两者都采用五指仿生布局——四指加可对握拇指——其运动学足够接近人手,可在标准面向人类的工装中执行工具操作任务。 OmniHand 2025 的六个被动 DoF 采用差动耦合:腱力会在所有接触到物体的指段间自然分布,贴合物体几何形状,无需控制器指定每个关节角度。这种欠驱动降低控制复杂度,并提高对不规则物体的抓取可靠性。对于需要精度的关节——拇指对握和食指远端指骨——独立驱动自由度提供了捏取硬币或 USB 接头等小物体所需的准确性。 OmniHand 3 Ultra-T 对全部 22 个主动手指 DoF 采用全腱驱动架构,并配有独立 3-DoF 腕部(55° 俯仰、40° 偏航)。腱驱动通过柔性腱绕滑轮传递驱动力,而非在每个关节放置电机,因此可把更多 DoF 塞进更轻的手中。公司声称 10:1 负载重量比(500 g 手部,最高 5 kg 有效载荷)对 25-DoF 设计来说非常突出;尚无独立第三方测试验证这一比例,低于 0.3 秒的整手响应时间声明同样没有披露测试协议。 OmniHand 产品线的触觉传感呈现清晰代际演进。OmniHand 2025(带触觉选项)使用 400+ 个力敏接触点,分辨率 0.1 N,覆盖 0–20 N 感知范围。Pro 升级到多模态方案:3 轴指尖力传感(捕捉法向压力和切向剪切力)、1 轴掌心力,以及 150+ 个总触点,分辨率 0.01 N,范围 0–50 N,最大无损力 1000 N。Ultra-T 声称具备「全手 3D 触觉传感」和集成掌心摄像头,但这一代没有公开触觉点数量、分辨率或感知范围。作为参照,人手掌面和手指约有 17,000+ 个机械感受器,这意味着即便 Pro 捕获的接触信息也只是生物密度的一小部分。[CE003, CE004, CE005, CE006, CE008, CE011]
| 手 | 总 DoF | 主动 DoF | 重量(g) | 最大指尖力(N) | 负载能力(kg) | 触觉点 | 驱动 / 传动 | 价格(USD 约) |
|---|---|---|---|---|---|---|---|---|
| 人手(参考) | ~27 | ~27 | ~400 | >90 每指 | >50 整只手 | 17,000+ 机械感受器 | 腱 / 肌肉 | N/A |
| OmniHand 2025 (Agilink) | 16 | 10 | ≤500 | 5 | 2(最大,掌心向上) | 400+(可选) | 连杆 / 电机 + 齿轮 | $1,350(标价)/ 约 $4,420(经销商) |
| OmniHand Pro 2025(Agilink) | 19 | 12 | ≤750 | 20 | 35(掌心向上) | 150+(多模态) | 连杆 / 电机 + 丝杆 | $12,700 |
| OmniHand 3 Ultra-T(Agilink,已宣布) | 25(22+3 腕部) | 22 | ~500 | 未披露 | ~5(10:1 比例,公司说法) | 全手 3D(公司说法) | 腱绳驱动 | 未披露 |
| LEAP Hand(CMU) | 16 | 16 | 940 | 未披露 | ~2 | 是(已集成) | 伺服电机 | $2,000 |
| Allegro Hand V4(Wonik Robotics 产品) | 16 | 16 | ~800 | 未披露 | 未披露 | 标配无 | 电机 + 连杆 | ~$15,000+ |
| XHAND1(Robotera) | 12 | 12 | 1,100 | 未披露 | 25 | 是 | 齿轮驱动力控关节 | $14,000 |
人手和 OmniHand 3 Ultra-T 的数据都不完整——人手规格是生物学参照值;Ultra-T 规格来自 Agibot 公司公告,未经独立验证。LEAP、Allegro 和 XHAND1 的对比数据来自产品页和第三方指南。「公司说法」表示公司披露但未验证的数值。
[CE001, CE004, CE009, CE012, CE014, CE017]5.3 软件栈——SDK、CANFD、ROS 2 与 AI 平台
Agibot(母公司)在 GitHub 以 Mulan PSL v2 许可证开源 OmniHand 2025 SDK。SDK 提供 Python 3.10+ 绑定和 C++ API,覆盖关节位置、力矩、速度和触觉控制模式。官方唯一支持的操作系统是 Ubuntu 22.04(x86_64),并使用 CMake 3.24+ 和 gcc 11.4+。预编译 Python wheel(omnihand_2025_py-0.8.0)简化安装。CANFD 通信通过 ZLG USBCANFD 系列适配器完成(推荐 USBCANFD-100U-mini、USBCANFD-100U 或 USBCANFD-200U)。仅支持 Ubuntu、仅支持 x86 的约束,把部署限制在工作站或服务器场景;公司没有记录 ARM-SoC、嵌入式或边缘部署路径。 OmniHand 2025 和 Pro 2025 SDK 均包含 URDF 模型,可集成进仿真环境(Genie Sim 3.0、NVIDIA Isaac Sim),并提供 ROS 2 驱动包,接入更广泛的机器人中间件生态。SDK 提供五种控制模式——Position/Angle(P-A)、Torque、Hybrid P+Torque(PtA)、Velocity 和 Tactile——足以支持遥操作和自主闭环操作。 平台层面,Agibot 的 Genie Studio 提供全栈开发环境,覆盖数据采集、基础模型训练(支持 RDT、Pi0、OpenVLA、GR00T 以及 AgiBot 自有 Go-1 模型)、仿真和一键部署。Genie Studio 处理来自多个机器人本体和末端执行器的跨设备数据采集;Agibot 称新场景微调周期为 48 GPU-days。托管在 HuggingFace(agibot-world 组织)的 AgiBot World 被描述为一个包含 1M+ 条轨迹、覆盖五个真实世界领域 217 项任务的数据集;该数据集使用带视觉触觉传感器和 6-DoF 灵巧手的硬件,与 OmniHand 产品族一致。截至报告运行日期,Agilink 或 Agibot 尚未公开 SDK 生命周期、版本节奏和长期维护承诺。[CE023, CE024, CE025, CE026, CE029, CE030]
| 层级 / 组件 | 系统角色 | 关键依赖 | 风险 / 缺口 |
|---|---|---|---|
| 机械结构(PA + 硅胶) | 手指、手掌和腕部实体结构;容纳执行器和传感器 | 供应商级 PA12 尼龙,以及指尖用 EcoFlex / 硅胶 | 材料国产与进口占比未披露;硅胶在反复受力下的耐久性未量化 |
| 执行机构(电机 + 齿轮 / 电机 + 丝杆) | 驱动 10–12 个主动 DoF;产生指尖力 | 电机供应商未披露;齿轮和丝杆质量决定重复精度 | 存在单一电机来源风险;Pro 的 0.02 mm 重复精度仅由公司验证 |
| 腱绳传动(OmniHand 3) | 通过绕过滑轮的柔性腱绳,把电机扭矩传到手指关节 | 腱绳材料、走线和预张紧质量 | 腱绳磨损和寿命未公开测试;更换周期未披露 |
| 触觉传感器阵列 | 检测接触力、剪切力和滑移事件,用于闭环控制 | 传感器厂商和集成接口未披露 | 400 点(O10)和 150 点(O12)密度远低于人手触觉分辨率;Ultra-T 声称「全 3D」,但未给出规格 |
| CANFD 通信(1 kHz) | 控制器与灵巧手之间的实时关节指令 / 反馈回路 | ZLG USBCANFD 适配器(USB-to-CANFD 桥接);依赖单一厂商 | ZLG 专用驱动可能限制主机可移植性;未记录备用方案 |
| OmniHand SDK(Python / C++) | 面向位置、力矩、速度、混合控制和触觉控制的开发者 API | Ubuntu 22.04 x86_64;CMake 3.24+;Mulan PSL v2 开源许可证 | 不支持 ARM / 边缘部署;SDK 0.8.0 版本显示成熟度仍在 1.0 之前 |
| ROS 2 驱动 | 面向科研和工业机器人栈的中间件集成 | 通过 AimRT 中间件兼容 ROS 2 Humble / Jazzy | ROS 2 版本兼容矩阵未发布;驱动生命周期承诺未披露 |
层级描述综合自官方产品页、GitHub SDK 文档和 robotscanada.com 技术指南。依赖和风险评估是基于行业常规的分析师推断;我们未审阅内部供应链或组件审计材料。
[CE003, CE006, CE007, CE023, CE024, CE025]端到端工作流展示机器人客户如何集成 OmniHand、生成训练数据,并借助 Agibot 的 AI 平台迭代到自主执行任务。
[CE006, CE023, CE029, CE030]5.4 部署证据与用例验证
与多数早期硬件拆分公司相比,OmniHand 产品族的商业部署证据更充分。Agibot 将发布价 ¥9,800 的 OmniHand 2025 描述为首个低于 ¥10,000 的高 DoF 手;它被纳入 AgiBot World 训练数据集,后者覆盖工厂、物流、家庭、酒店和商业五个真实世界领域,至少间接证明其在数据采集工作流中被反复功能性使用。 OmniHand Pro 2025 最强的部署证据来自 G2 工业仿人机器人发布活动:配备 OmniHand Pro 的 G2 执行了物流包裹分拣(不同尺寸、形状和材料)和精密装配任务(RAM 模块插入、安全带锁芯装配)。这些并非经典意义上的受控演示:它们处理的是未排序的混合 SKU 包裹,而不是预先分好的相同物体。Agibot 还报告两条产品线累计交付超过 8,000 只灵巧手,公司 2026 年 3 月达到 10,000 台机器人里程碑则覆盖完整产品组合。没有客户具名案例、部署规模细节或现场故障数据发布,因此难以评估复购率或真实世界可靠性。 Robotera 的 XHAND1(12 DoF、$14,000)是直接产品市场竞争者,同样面向仿人机器人集成。Carnegie Mellon 的 LEAP Hand(16 DoF、$2,000、舵机驱动、约 2 kg 载荷)代表开源研究可及档——载荷和传感能力明显较弱,但价格约为 OmniHand 2025 的一半,并且已有学术引用基础(LEAP Hand 论文发表于 RSS 2023)。Allegro Hand V4(Wonik Robotics)是成熟研究标准,拥有 16 DoF 和丰富 ROS 生态,但价格 $15,000+,机械能力弱于 OmniHand Pro。Fourier GR-2 集成 12-DoF 手(每只手 6 个阵列式触觉传感器),Shadow Dexterous Hand 仍是实验室金标准,24 DoF、价格约 $130,000+。Agilink 的 OmniHand Pro 在中端价位上以 35 kg 负载能力超过所有具名替代品,但尚无第三方对比基准。[CE027, CE028, CE031, CE032, CE033, CE038]
| 用户细分 | 当前工作流(无 OmniHand) | OmniHand 方案 | 可量化收益(按证据) | 局限 / 待解问题 |
|---|---|---|---|---|
| 工业物流 / 包裹分拣 | 人工拣选,或按包裹类型切换固定夹爪 | G2 移动操作臂搭载 OmniHand Pro;单一末端执行器可处理多样 SKU,无需换型 | G2+OmniHand Pro 在 2025 年 G2 发布会上现场演示混合包裹分拣 | 未披露吞吐率、错误率或多班次可靠性数据 |
| 精密装配(电子、汽车) | RAM 插装、安全带锁芯、细间距零件由人工操作 | OmniHand Pro 以 0.02 mm 重复精度和 20 N 指尖力处理组件 | G2 发布会演示了 RAM 模块插装和安全带锁装配 | 未披露节拍、缺陷率,或与人工操作速度的对比 |
| 互动服务 / HRI(人机交互) | 专用服务机器人手缺少触觉反馈或安全设计 | OmniHand 2025 配有防夹设计和手背触摸感知 | AgiBot A2 演示包括拆包、倒液、递交、手势;未见用户研究数据 | 未披露正式 HRI 安全认证(如 ISO/TS 15066 协作机器人标准) |
| 具身 AI 训练 / 研究 | 研究级灵巧手价格高($14K–$130K+),限制数据集规模 | OmniHand 2025 定价 ¥9,800,可支撑大规模遥操作数据采集;AgiBot World 已采用 | AgiBot World:100 万+ 轨迹、217 项任务;数据集使用 OmniHand 硬件 | 学术社区获取需要处理进口物流和 CANFD 适配器采购 |
| 高校与教育实验室 | 可及选择主要是 Allegro V4(约 $15K)或 LEAP Hand($2K,载荷有限) | OmniHand 2025 以 ¥9,800 / ~$4,420 的价格和开放 SDK,填补 LEAP 与 Pro 之间的空档 | DoF 与 Allegro 接近,价格约为其 30%;SDK 成熟度低于 Allegro 十年的 ROS 支持 | 未宣布面向高校的授权、仿真预集成包或教育折扣 |
部署证据来自公司演示、产品公告和已发布的数据集文档。目前没有独立客户验证、现场可靠性报告或吞吐基准被公开披露。
[CE007, CE008, CE027, CE028, CE030, CE031]关键供应商、平台和监管依赖的 DAG,决定 Agilink 能否制造、支持并放大 OmniHand 产品线。
[CE024, CE025, CE041]5.5 差异化、技术风险与尽调缺口
Agilink 的核心技术差异化论点包括:(1)三档硬件梯度,在同一生态中覆盖研究价格到工业能力;(2)量产血统——OmniHand 2025 在拆分前已由 Agibot 商业供货,相比纸面阶段竞争者具备制造成熟度优势;(3)通过 Genie Studio 和 AgiBot World 集成 AI 训练基础设施,形成闭环数据飞轮,理论上可随时间改善操作策略;(4)在认可度较高的中国开源许可证下开源 SDK,降低学术和 OEM 客户集成摩擦。 与这些优势相对,本章指出五项重大技术风险。第一,OmniHand 3 Ultra-T 的旗舰规格——10:1 负载重量比、低于 0.3 秒响应、全手 3D 触觉——完全依赖 Agibot 的 APC 2026 新闻稿,没有独立复现、客户引述或详细测试协议;机器人硬件中的超常声明,经常无法在独立评估中以完整精度复现。第二,相比生物密度,各档触觉传感成熟度有限:人手约有 17,000+ 个机械感受器,而 OmniHand 2025 为 400 点(可选)、Pro 为 150 点阵列;Ultra-T 的全手 3D 声明缺少量化规格。第三,SDK 锁定 Ubuntu 22.04 x86_64,未记录 ARM / edge 路径,会给使用 Raspberry Pi、NVIDIA Jetson 或 Qualcomm 硬件的机器人团队带来集成摩擦。第四,任何 OmniHand 版本都没有公开 MTBF、耐久循环次数(类似 OmniPicker 3 的 1,000,000 次循环评级)、保修条款或现场故障率数据,使部件寿命和总拥有成本无法分析。第五,agilink.cn 域名未投入运营,产品仍托管在 agibot.com,引发关于品牌独立性、IP 归属以及拆分时具体转移了哪些 IP 的问题;这些对任何授权、赔偿或收购情景都很重要。[CE017, CE018, CE019, CE020, CE022, CE025]
| 日期 / 时段 | 里程碑 / 功能 | 状态 | 含义 | 来源 |
|---|---|---|---|---|
| 2025 年 8 月 | OmniHand 2025 系列发布(Agile + Pro 版本) | 已发布——可商购 | 首个 ¥10,000 以下高 DoF 灵巧手;Pro 面向工业买家定价 $12,700 | Agibot 官方;China Youth International |
| 2025 年 10 月 | OmniHand Pro 随 Agibot G2 工业人形机器人发布 | 已发布——在产线场景演示 | 在物流和装配现场演示中验证 35 kg 载荷和 0.02 mm 重复精度 | Humanoids Daily;Interesting Engineering(第 1–4 章背景) |
| 2025 年 12 月(推断) | OmniHand SDK v0.8.0 发布在 GitHub(Mulan PSL v2) | 已发布——1.0 前版本,仍在活跃更新 | Python 3.10+ / C++ API;仅支持 Ubuntu 22.04;通过 ZLG 适配器接 CANFD | GitHub AgibotTech/Omnihand-2025-SDK |
| 2026 年 2 月 | OmniHand 2025 和 Pro 的技术规格 PDF 均更新 | 已发布——文档刷新 | 暗示剥离后产品支持仍在持续,规格趋于稳定 | Robots Canada OmniHand 技术指南 |
| 2026 年 4 月 17 日 | OmniHand 3 Ultra-T、3 Lite、OmniPicker 3 在 APC 2026 发布 | 已宣布——未披露供货日期或定价 | 22+3 DoF 腱绳驱动架构;10:1 载荷比(公司说法);掌心摄像头 | Agibot APC 2026 新闻稿 |
| 2026 年(持续) | Agilink 产品品牌从 agibot.com 迁移到独立域名 | 未完成——截至 2026 年 6 月,agilink.cn 域名尚未启用 | 引出 IP 归属、客户支持连续性和品牌清晰度问题 | 分析师推断;没有公开迁移计划公告 |
SDK 0.8.0 的发布日期是根据 GitHub 提交历史语境推断;所审阅来源没有直接说明确切发布日期。品牌迁移一行是分析师基于 agilink.cn 域名未运行作出的推断。
[CE022, CE023, CE025, CE027, CE039, CE041]用 1–5 分定性量表,从六个能力维度给 OmniHand 2025、Pro 和 Ultra-T 打分,凸显产品线内部成熟度不均。
评分是分析师基于已发布规格和证据的定性判断(1=低 / 缺失,5=高 / 成熟);不是标准化基准测试结果。Ultra-T 分数反映仅发布状态。LEAP Hand 作为开源研究参照纳入。
[CE017, CE022, CE031, CE043]5.6 展示材料
06客户
6.1 客户基础与部署版图
Agilink 的 go-to-market 与母公司 Agibot 的全栈机器人平台不可分割。OmniHand 灵巧手作为标准操作末端执行器嵌入每一台 Agibot G2 Series 工业仿人机器人;截至 2026 年 3 月,Agibot 已在 17 个国家累计出货 10,000 台仿人机器人,仅 2025 年出货 5,100 台(Omdia,Humanoid Index 引用,按出货量全球第一)。Agibot 宣布的八类商业应用覆盖工业制造、物流分拣、零售导引、酒店服务、安防巡逻、数据采集、研究和清洁。每次部署都隐含使用 OmniHand 的触觉驱动操作能力,但 Agilink 的独立收入、向 Agibot 的单价,以及第三方灵巧手销售仍完全未披露。唯一拥有生产级指标的已确认外部客户是全球智能设备 ODM 制造商 Longcheer Technology。Agibot 的 APC 2026「Seven Solutions」新闻材料中还出现了另外十二个具名企业账户(HT-Tech、Intel、SAIC-GM、FULIN Precision、Damon、Xiaonan Intelligence、Anta、Haidilao、Guangzhou Metro、Digital China、State Grid、Hongqiao Airport),但没有任何一家通过自身新闻稿独立确认与 Agibot 的关系。国际合作伙伴(Genting Malaysia、Hyperscale Data、Singtel、SIR Spa、Whale Cloud)在 APC 2026 和 APC Hong Kong 2026 宣布,但截至本报告运行日期,尚未进展为有文件记录的生产部署。 [CU027, CU028, CU029, CU026, CU031, CU039]
| 细分 | 主要客户 / 合作伙伴 | 部署用例(OmniHand / G2) | 规模 / 阶段 | 证据质量 | 尽调缺口 |
|---|---|---|---|---|---|
| 消费电子制造 | Longcheer Technology(南昌,江西) | 平板 MMIT 装配、精密上下料 | 生产(24/7,310 UPH) | 高——3 个交叉印证来源,具名高管引述 | 合同条款、单价、续约情况未披露 |
| 半导体搬运 | HT-Tech;Intel | 精密芯片上下料,<0.001% 掉片率 | 生产(声称已规模化,未独立确认) | 低——仅单一 Agibot 来源 | HT-Tech 或 Intel 未发布独立新闻稿 |
| 汽车制造 | SAIC-GM;FULIN Precision | 电池电芯搬运;标准化容器转运 | 部署(规模未说明) | 低——仅单一 Agibot 来源 | 未独立确认;未披露设备数量 |
| 物流与履约 | Damon;Xiaonan Intelligence | 包裹分拣;软包分拣 | 试点 / 早期部署 | 低——仅单一 Agibot 来源 | 达到人工吞吐的比例;是否投产未确认 |
| 零售与服务 | 客户:Anta Sports;Haidilao;Guangzhou Metro | 顾客引导、排队、咨询解答 | 试点 / 有限部署 | 低——仅 Agibot 来源指标 | 客流 / 转化率提升未验证;无客户声明 |
| 酒店 / 娱乐 | Genting Malaysia(MOU) | 主题公园、酒店礼宾、现场演出 | 2026 年 4 月签署 MOU——尚未部署 | 无——仅 MOU 阶段 | 部署时间表和合同范围未披露 |
| 数据采集 / 北美 | Hyperscale Data / Omnipresent Robotics | 按美国工业标准加固机器人,采集遥操作数据 | 已宣布合作——计划建设 100K sqft 设施 | 无——仅公告阶段 | 截至 2026 年 6 月,设施运营状态和时间表未知 |
| 独立 OmniHand OEM | 未公开识别 | 第三方机器人 OEM 集成 | 未记录 / 零证据 | None | 关键缺口:没有任何独立 OmniHand OEM 客户证据 |
证据质量评级依据交叉印证深度,而非 Agibot 自身性能说法。「生产」要求有外部印证的连续运营记录;「试点」指单一来源说法且没有具名客户确认。合作伙伴 / MOU 行代表承诺,不代表实际部署。
[CU001, CU010, CU011, CU012, CU013, CU014]图中梳理 OmniHand 触达终端用户的四类客户关系,从已确认的 Agibot 内部嵌入式需求,到尚无文档支撑的独立 OEM 渠道。
节点位置和流向是基于新闻稿及合作伙伴公告的分析推断;没有披露组织架构图或渠道合同结构。
[CU029, CU031, CU023, CU025, CU039]6.2 具名客户证据:嵌入 Agibot 的部署
Longcheer Technology 是 Agilink / Agibot 的旗舰生产规模客户。配备 OmniHand 的 Agibot G2 机器人部署在江西南昌 Longcheer 平板组装工厂(MMIT 工位),在四个月项目爬坡后,通过 Genie Sim 3.0 数字孪生平台用 36 小时完成集成。独立佐证指标包括 310 UPH 吞吐、99.9%+ 动作成功率,以及 8 小时连续运行中完成 2,283 次零错误任务;系统支持 24/7 运行,停机损失低于 4%。Longcheer 机器人事业部总经理 Li Long 在官方 PRNewswire 新闻稿中提供具名引述。Agibot 目标在 2026 年 Q3 前把 Longcheer 部署扩大到 100 台机器人。除 Longcheer 外,APC 2026「Seven Solutions」演示还列出其他部署:HT-Tech(最高 1M 芯片 / 天、掉落率 <0.001%)、Intel(100,000 芯片 / 天)、SAIC-GM(电芯搬运、亚秒级延迟)、FULIN Precision(容器运输)、Damon(达到人工分拣吞吐的 70%、99% 二次通过)、Xiaonan Intelligence(435 件 / 小时、人工成本降低 50%)、Anta 门店(客流提升 20%)、Haidilao 餐厅、Guangzhou Metro(成本降低 13%、80% 自助解决率)和 State Grid(巡检效率数倍提升)。上述段落中除 Longcheer 以外的所有指标都只来自 Agibot 自有营销材料;没有独立客户署名新闻稿佐证这些部署。 [CU001, CU002, CU003, CU004, CU005, CU006]
| 客户 / 合作伙伴 | 细分 | 部署用例 | 生产 / 试点 | 关键结果 / 指标 | 证据局限 |
|---|---|---|---|---|---|
| Longcheer Technology | 消费电子 ODM | Agibot G2 + OmniHand 执行平板 MMIT 装配 | 生产——24/7 连续运行 | 310 UPH;99.9%+ 成功率;8 小时完成 2,283 项任务;36 小时集成 | 所有指标均来自 Agibot;无独立审计;Li Long 引述可交叉印证,但仍出自 Agibot 新闻稿 |
| HT-Tech | 半导体 | 精密芯片上下料 | 部署——规模未确认 | 最高 1,000,000 颗芯片 / 天,掉片率 <0.001% | 仅在 Agibot Seven Solutions 中具名;无 HT-Tech 独立声明 |
| Intel | 半导体(跨国公司) | 100,000 颗芯片 / 天的芯片搬运 | 部署——规模未确认 | 声称 100,000 颗芯片 / 天 | Intel 未独立确认;考虑到 Intel 采购标准,这是异常强的说法 |
| SAIC-GM | 汽车(合资公司) | 以亚秒级感知延迟搬运电池电芯 | 部署——规模未确认 | 接近连续吞吐;亚秒级延迟 | SAIC-GM 被具名但无独立新闻稿;汽车合资公司理论上可核验,但尚未完成 |
| Damon | 物流 | 包裹分拣;使用「测试和部署」表述 | 试点 / 早期部署 | 达到人工水平 70% 吞吐;二次分拣成功率 99% | 「测试和部署」措辞暗示尚未完全达到生产级 |
| Anta Sports | 零售——服饰 | 顾客引导与零售协助 | 试点 | 客流提升 20%;转化率取得两位数增长 | 无 Anta 来源声明;指标周期和门店数量不清 |
| Guangzhou Metro | 公共基础设施 | 乘客咨询与引导(80% 自助) | 部署 | 运营成本下降 13%;80% 咨询由自助完成 | 无 Guangzhou Metro 声明;成本下降口径未知 |
| State Grid | 能源 / 基础设施 | 安防巡逻与巡检 | 部署 | 巡检效率提升「多倍」;维护成本下降 | 指标极其含糊;未披露设备数量或合同范围 |
| Genting Malaysia Bhd | 酒店 / 娱乐 | 主题公园、酒店、现场演出集成 | 仅 MOU(2026 年 4 月 17 日) | 计划在 2027 年举办 Robotics Gala Performance | 不是部署;MOU 可能无法在 12 个月内转化为合同 |
| Hyperscale Data(Omnipresent Robotics) | 数据采集 / 北美 | 按美国工业标准加固机器人;遥操作数据 | 合作 / 设施规划中 | 密歇根 100,000 sqft 设施;目标 500 个技术岗位 | 仅公告;未确认部署;设施建设时间表未知 |
本表仅列出公开具名的客户 / 合作伙伴。证据质量按交叉印证深度评估;「生产」行要求至少两个独立来源。第 2–10 行的指标全部来自 Agibot APC 2026「Seven Solutions」材料或合作伙伴大会公告;没有一项得到具名实体独立确认。含有「未确认」或「规模未确认」的单元格表示缺乏独立验证,并非否认。
[CU001, CU002, CU003, CU004, CU005, CU008]Agibot G2 部署从发现到量产的路径,以 Longcheer 部署时间线作为有文档支撑的模板。
漏斗阶段和时间来自 Longcheer 部署案例。其他已点名部署没有披露集成时间线。
[CU001, CU005, CU006, CU007, CU027]6.3 独立 OmniHand 渠道与第三方市场
没有公开文件证据显示任何第三方 OEM、大学研究实验室或系统集成商,脱离打包的 Agibot G2 机器人,直接从 Agilink 购买 OmniHand 作为独立灵巧手产品。Agibot 在 CES 2026 披露的「Powered by AGIBOT」定制计划允许合作伙伴配置机器人硬件和软件,意味着 OmniHand 可能通过这一渠道进入第三方机器人平台,但没有公开具名确认这一路径的 OEM 客户。Agilink 在 store.agibot.com.cn 的电商列表和零售商页面(Botonly、Robots Canada)显示其可零售购买,但销量和客户身份未知。Pandaily 报道 Agilink 到 2026 年初已出货超过 8,000 只灵巧手,但文章正文不可访问;若该数字准确,意味着在当时估算约 3,000 台 G2 机队之外存在显著独立渠道,但来源缺口使该指标不能作为尽调依据。可比竞争生态提供了参照:RobotEra(XHAND1)披露已通过 China Post 和 SF Group 在 10+ 个中心部署物流场景,RightHand Robotics 则具名 PALTAC(日本)和 Apologistics(荷兰)为生产客户。二者都展示了客户披露门槛,而 Agilink 的独立手部业务尚未跨过这道门槛。 [CU030, CU031, CU032, CU033, CU034, CU039]
| 指标 | 数值 | 日期 | 来源 | 置信度 | 含义 | 缺失分母 / 注意事项 |
|---|---|---|---|---|---|---|
| Agibot 人形机器人累计出货量 | 10,000 | 2026 年 3 月 | Agibot 新闻稿;Humanoids Daily | 中 | OmniHand 默认嵌入所有已出货 G2 机型 | 未披露机型层级(G2 vs A2 vs X2)拆分 |
| 2025 年出货量(全球排名第 1) | 5,100 | 2025 年底 | Humanoid Index 引用 Omdia | 中低 | 按销量看处于市场领先,但 Unitree 有异议 | Omdia 方法不透明;Unitree 称为 5,500 |
| OmniHand 灵巧手出货量(Agilink) | 8,000+ | Q1 2026 | Pandaily(仅标题;正文不可访问) | 低 | 暗示有独立灵巧手销售;可能包含嵌入 G2 的设备 | 文章正文返回为空;该数字无法独立验证 |
| Longcheer 到 2026 年 Q3 的 G2 机器人目标 | 100 | 目标:2026 年 Q3 | Agibot 官方;PRNewswire | 低(前瞻指引) | 仅单一客户规模目标 | 截至 2026 年 6 月,前瞻指引未确认 |
| Agibot 机队覆盖国家数 | 17 | 2026 年 4 月 | Agibot APC 2026 | 中低 | 声称已有全球分布,但 17 个国家中 16 个缺少客户名称 | 「机队覆盖」可能包括演示机,而非签约部署 |
| Longcheer 产线吞吐(UPH) | 310 | 2026 年 4 月 | PRNewswire;Agibot 官方;CnTechPost | 高 | 唯一获得独立交叉印证的量化部署指标 | 未经独立审计;基于公司 + 媒体报道 |
除标注为第三方(Omdia、Pandaily)的指标外,所有体量指标均来自 Agibot 自报。前瞻指引指标标为「低」置信度。8,000+ 灵巧手数字仅见标题,无法从可访问正文验证。
[CU027, CU028, CU030, CU006, CU026, CU002]6.4 部署质量与客户证据成熟度
Agibot 生态中的客户证据质量差异很大。Longcheer Technology 是最高质量证据:指标得到独立佐证,官方新闻稿中有具名客户高管引述,并且四个月部署时间线具备可验证里程碑。Seven Solutions 演示中的另外十二项具名企业部署,证据质量为中到低:结果被量化(芯片掉落率、成本下降、吞吐指标),但证据来源完全是 Agibot 自有 APC 2026 文件,没有链接客户案例、独立新闻稿或分析师确认。Morgan Stanley 对中国企业决策者的 AlphaWise 调查显示,只有 23% 目前对仿人机器人产品满意,灵巧度、功能和成本被列为主要障碍;截至 2026 年初,仅约 10% 受访者正在主动评估或试点仿人机器人。这一全行业不满背景并不直接反驳 Agibot 的部署声明,但它界定了早期多数市场的采用挑战。国际合作伙伴公告(Genting Malaysia MOU、Hyperscale Data Michigan facility、Singtel RaaS)仍处于 MOU / 公告阶段,尚非部署。任何已识别 Agilink / Agibot 客户都没有披露 NRR、流失、续约率或复购数据。 [CU001, CU008, CU021, CU022, CU023, CU024]
| 指标 | 数值 / 状态 | 细分 | 置信度 | 尽调问题 |
|---|---|---|---|---|
| 净收入留存率(NRR) | 未披露 | 所有细分市场 | N/A | 向 Agilink/Agibot 投资者关系索取 NRR 数据;该数据对独立剥离主体估值至关重要 |
| 流失率(已部署机器人召回或合同终止) | 未披露 | 所有细分市场 | N/A | 确认 Longcheer 是否在初始部署阶段后扩展或续约 |
| 当前人形机器人客户满意度(Morgan Stanley 行业调查) | 23% 满意 | 中国企业高管整体样本 | 中 — 投行调查,约 100+ 名受访者 | 获取 Agibot 特定客户满意度数据,以区别于行业平均值 |
| 机型切换时设备复用率(Agibot 声称) | 95% | 消费电子制造 | 低 — 公司声称,未审计 | 向 Longcheer 索取切换测试协议和独立验证 |
| 已确认重复 / 扩单订单的活跃部署 | 1(Longcheer — 100 台机器人目标) | 消费电子 | 中 — 仅前瞻指引 | 确认 Q3 2026 扩产是否完成,以及是否已签约更多场地 |
所有留存和满意度指标要么缺失(未披露),要么来自 Agibot 自身说法。Morgan Stanley 调查覆盖的是更广泛行业,不是 Agilink/Agibot 个体。null 值表示该指标公开不可得,不是零。
[CU035, CU036, CU006, CU039]从来源独立性、指标具体度、生产成熟度和留存可见度四个维度,对比四类客户 / 合作伙伴层级的部署证据质量。评级基于可获得公开证据,属定性判断(强 / 中 / 弱 / 无)。
评级是分析师定性判断。来源独立性为“无”表示唯一证据来源是 Agibot 自有新闻材料。RightHand 同业基准行仅用于校准。
[CU001, CU034, CU035, CU038, CU039]6.5 集中度、依赖性与披露风险
客户集中风险很高:Longcheer Technology 是唯一有独立证据的生产规模具名客户,使 Agilink 已证明的外部需求实际上成为单一账户故事。更深的风险是结构性母公司依赖:所有已识别部署都针对 Agibot 完整 G2 / A2 机器人,Agilink 的收入、转移定价以及与 Agibot 的公平交易关系完全未披露。如果 Agilink 主要是 Agibot 的 captive supplier——证据确实如此暗示——它的客户基础并不真正独立,作为独立拆分公司的估值溢价取决于未来第三方需求,而这些需求尚未公开兑现。地缘政治暴露带来二阶风险:2026 年 3 月由 Cotton 和 Schumer 参议员提出的 American Security Robotics Act 将禁止美国联邦政府采购中国公司制造的仿人机器人;Unitree 受争议的出货声明也为整个行业基于销量的营销带来可信度风险。另一个问题是,Agibot 声称其在消费电子制造中实现了「全球首个」大规模具身 AI 部署,这一表述带有狭窄限定(生产规模的精密电子),在 Amazon 的 Agility Robotics 和 BMW Spartanburg 工厂的 Figure AI 等可比工厂试点面前,可能经不起审视。 [CU037, CU038, CU039, CU040]
| 风险 / 驱动因素 | 类型 | 影响 | 尽调路径 |
|---|---|---|---|
| 单一具名生产客户(Longcheer)主导可验证证据 | 客户集中度 | 高 — 整个外部部署叙事压在一个账户上 | 找到并独立验证至少 2 个额外生产客户;要求安排客户访谈 |
| Agilink 收入 = Agibot 转移定价(内部供货) | 母公司依赖 | 高 — Agilink 独立财务和公允定价未知 | 索取 Agilink 独立 P&L;确认转移定价方法和审计独立性 |
| 未公开识别出独立 OmniHand OEM 客户 | 市场开拓风险 | 高 — 第三方灵巧手市场可能尚未形成规模 | 监测 Agilink 渠道公告;检索学术和分销商材料中的 OmniHand 采购线索 |
| American Security Robotics Act(美国联邦采购禁令提案) | 地缘政治 / 监管风险 | 中高 — 将阻断美国政府客户,也可能吓退美国企业线索 | 监测法案进展;评估北美计划中企业客户与政府客户的拆分 |
| Intel 和 SAIC-GM 被点名但缺少独立确认 | 披露可信度风险 | 中 — 若这些账户被错误归因或只是试点,头部客户名单被高估 | 索取 Intel 或 SAIC-GM 新闻稿,或能确认关系的法律 / 合同材料 |
| “全球首个”消费电子部署说法 | 营销可信度风险 | 中 — Agility Robotics(Amazon)和 Figure AI(BMW)已先于 Agibot 进入类似工厂环境 | 核验限定范围(“量产规模精密电子”),并判断竞争对手更早部署是否符合相同标准 |
影响评级是尽调判断,不是市场预测。“高”表示该风险会实质影响 Agilink 独立估值或客户管线规模测算。缓释动作按信息需求紧迫度排序。
[CU039, CU040, CU037, CU038, CU011, CU001]条形图展示截至 2026 年 6 月,Agilink / Agibot 命名账户按证据质量层级的分布。Longcheer 是唯一有独立佐证、达到生产级证据的账户;其余命名企业账户要么只有 Agibot 来源,要么仍处于 MOU / 合作阶段。
账户数基于 Agibot APC 2026 及相关新闻稿公开点名的客户 / 合作伙伴。OEM 渠道值为 0,反映截至运行日期没有公开识别出的独立 OmniHand 客户。
[CU001, CU039, CU038, CU021]6.6 展示材料
07风险
7.1 地缘政治与监管风险
Agilink 的中国背景带来集中的地缘政治风险,横跨美国立法、多边出口管制和欧洲产品安全法。2026 年 3 月 26 日,美国参议员 Tom Cotton 和 Chuck Schumer 提出 American Security Robotics Act(ASRA),这是一项两党法案,拟禁止美国联邦政府机构购买或运营中国公司制造的仿人机器人。US News and World Report 报道称,Agibot 和 Unitree 在法案文本中被明确点名,众议员 Stefanik 也提出了配套众议院法案。法案尚未通过,但两党联署以及 DJI / Huawei 等中国硬件禁令先例,会对企业采购形成可信的寒蝉效应,尤其是在美国国防相关制造场景。 ASRA 之外,Bureau of Industry and Security(BIS)在 2025 年 1 月发布先进计算 IC 规则,限制向总部位于 Country Group D:5(包括中国)的实体出口高端芯片。BIS 官网确认,BIS 已将 Authorized IC Designer 时间线延长至 2026 年 12 月 31 日。Agilink 灵巧手内置 AI 推理处理器,但具体分类没有公开披露;因此,限制只能表述为「可能受」EAR 管制,而非已确认。若 Agilink 的嵌入式推理芯片组出口需要 BIS 许可证,国际订单的供应链连续性将面临结构性不确定。 EU AI Act(Regulation 2024/1689,2026 年 8 月生效)将作为关键基础设施安全部件使用的 AI 系统归为高风险,要求在进入欧盟市场前完成合格评定、CE 标志和 NIST 对齐文档。在欧盟成员国产线上运行的灵巧手明确落入这一分类。Hill Dickinson 对仿人机器人法律的 2026 年分析还指出,GDPR 下的数据隐私——尤其是生物识别传感和云端触觉数据存储——是 Agilink 尚未公开处理的额外合规维度。 中国国内,MIIT 于 2026 年 2 月发布首个国家级仿人机器人标准体系。Agibot 联合创始人 Peng Zhihui 与 Unitree 的 Wang Xingxing 一同担任 MIIT 标准技术委员会副主任。这一位置让 Agibot / Agilink 能影响标准制定,但也带来监管俘获风险:如果 Agibot 关联规格嵌入专有接口,竞争对手可能发现标准本身成了护城河;反过来,Agilink 的竞争者也可能提出标准滥用异议,拖慢 Agilink 自身合规路径。 OSHA 已确认没有面向机器人行业的具体安全标准,ISO 10218-1:2011(工业机器人安全)也已正式撤销,尚无面向仿人机器人或灵巧手的替代标准。ISO/TS 15066:2016 覆盖协作机器人,但没有处理多 DoF 灵巧手的力限制或故障模式。这一监管缺口既造成产品责任模糊,也给任何随后引入标准的司法辖区带来先行者认证风险。[CR001, CR002, CR003, CR004, CR005, CR006]
| 风险 / 规则 | 司法辖区 | 状态(2026 年 6 月) | 可能性 | 严重度 | 缓释状态 | 剩余暴露 | 尽调路径 |
|---|---|---|---|---|---|---|---|
| American Security Robotics Act(ASRA)— 禁止中国人形机器人进入美国联邦采购 | 美国 | 已提交参议院 / 众议院;尚未通过 | 中 | 关键 | 未披露 | 高 — 即便未生效,也会对企业采购产生寒蝉效应 | 监测法案进展;索取 DJI 先例适用性的法律意见 |
| BIS 先进计算 IC 出口管制(EAR / Country Group D:5) | 美国 | 规则于 2025 年 1 月发布;IC 设计方时间线延至 2026 年 12 月 31 日 | 高 | 高 | 低 — 未披露具体芯片分类 | 高 — AI 推理硬件存在供应中断风险 | 披露嵌入式芯片组规格;向出口管制律师取得 EAR 分类意见 |
| EU AI Act 高风险 AI 分类(Regulation 2024/1689) | 欧盟 | 2026 年 8 月生效 | 高 | 高 | 低 — 未披露合规评估 | 高 — 取得 CE 标志前会阻断欧盟市场进入 | 启动合规评估;聘请欧盟公告机构;评估 GDPR 生物识别数据责任 |
| 中国《国家情报法》2017 — 强制配合国家义务 | 中国 / 全球 | 已生效 | 中 | 中 | 部分 — 存储在中国的数据可能可被调取 | 中 — 在国防相邻垂直行业造成客户感知和采购风险 | 向潜在企业买家披露数据架构和数据主权控制 |
| MIIT 人形机器人标准体系(2026 年 2 月)— 监管分化和 IP 锁定风险 | 中国 | 已发布;仍在制定 | 低 | 中 | 部分 — Agibot/Agilink 参与委员会 | 中 — 与 ISO/IEC 分化可能阻断互认 | 跟踪 ISO TC299 协调;为国际客户发布合规路线图 |
行按可能性 × 影响严重度排序。截至 2026 年 6 月;除非注明,所有立法事项均为提案,尚未生效。司法辖区覆盖的是主要出口市场风险,不是中国国内市场。
[CR001, CR002, CR003, CR004, CR005, CR006]六个风险领域按估算可能性(1-5)和影响(1-5)作图,并给出剩余严重度和缓释成熟度;单元格值为作者截至 2026 年 6 月的估算。
可能性和影响分数是作者基于已审阅证据给出的估算;并非来自正式风险模型。所有分数截至 2026 年 6 月。
[CR001, CR016, CR034, CR026, CR007]7.2 产品质量、安全标准缺口与可靠性风险
Agilink 的核心产品风险,是 IDC 和 Goldman Sachs 在 2026 年指出的行业首要技术瓶颈:sim-to-real 精度坍塌。IDC 2026 年 5 月商业化报告明确称,「仿真环境中 80-90% 的准确率,在真实场景中常常跌破 50%」;Taibo Financial 对 Goldman Sachs 2026 年 1 月中国调查的译文也佐证这一点,并补充称「大规模、高质量真实数据采集所需时间限制部署速度」。对于灵巧手产品,触觉传感和操作精度是核心价值主张,真实世界准确率 50% 或更低是品类层面的可信度风险,不只是产品工程问题。 唯一独立记录的生产部署位于 Longcheer Technology 南昌工厂,嵌入 OmniHand 的 Agibot G2 机器人报告达到 310 UPH 和 99.9% 成功率。Interesting Engineering 确认了这一部署,但指出它是在受控条件下的单一场址;能否扩展到多工厂、多班次、多产品制造环境仍未验证。Morgan Stanley 针对中国企业买家的 AlphaWise 调查(Humanoids Daily,2026 年报道)发现,受访 C-suite 中只有 23% 对当前仿人机器人产品满意,灵巧度和功能是主要抱怨——这是当前一代灵巧手市场接受度脆弱的直接证据。 另一个并行担忧是,Agilink 的产量声明未经审计。Humanoids Daily 报道称,「该行业生产数字未审计,无法独立验证」,并引用 Agibot(5,168 台)与 Unitree(5,500 台)围绕 2025 年仿人机器人出货冠军的公开争议。如果母公司 Agibot 自己的出货数据都存在争议,Agilink 来自 Agibot 来源材料的 8,000+ 累计灵巧手交付声明也同样难以独立佐证。 安全标准方面,OSHA 已确认机器人行业没有具体 OSHA 标准。历史上的工业机器人安全参考 ISO 10218-1:2011 已撤销。ISO/TS 15066:2016(协作机器人)存在,但它是技术规范,不是强制标准,也不覆盖灵巧多 DoF 手。NIST 的 AI 风险框架推动 AI 系统的测量科学和基于风险的治理,但没有约束力。三层缺失——没有 OSHA 规则、没有现行 ISO 标准、没有强制 NIST 要求——意味着 Agilink 面向的市场中,手部相关工伤的产品责任会依据一般产品责任法诉讼,而不是行业专属合规框架,损害赔偿暴露不可预测。[CR016, CR017, CR018, CR019, CR020, CR021]
| 失效模式 | 可能性 | 严重度 | 缓释成熟度 | 剩余暴露 | 未解决缺口 |
|---|---|---|---|---|---|
| 仿真到真实精度坍塌:80-90% 仿真精度 → 真实部署低于 50% | 高 | 高 | 部分 — Agibot-Agilink 数据飞轮在开发中 | 高 — 当前一代灵巧手可能低于客户预期 | OmniHand 3 系列未披露测试协议或真实世界精度规格 |
| 灵巧手缺少适用机器人安全标准(OSHA 缺口;ISO 10218 已撤回) | 高 | 高 | 无 — 未披露认证路径 | 高 — 任何 G7 部署都存在产品责任模糊 | 启动 ISO TC299 参与;发布力限和失效模式规格 |
| 单场地生产证明 — Longcheer Technology 是唯一带量化指标的部署 | 高 | 中 | 无 — 没有第二个确认生产场地 | 中 — 可复制性未验证 | 需要第二、第三个部署场地指标,才能达到投资级信心 |
| 未审计生产说法 — 出货数字未获独立验证 | 高 | 中 | 无 — 没有第三方审计 | 中 — 独角兽估值下存在财务虚高风险 | 要求第三方生产审计或 STAR Market 招股书披露 |
| 产品责任模糊 — 制造商、运营方、AI 提供方责任划分 | 中 | 高 | 未披露 | 高 — 中国或 G7 司法辖区尚无灵巧手伤害先例 | 取得产品责任保险;发布合同责任分配框架 |
按严重度排序。成熟度评级:无 = 未披露计划;部分 = 有关注证据但未发布协议;充分 = 有记录流程或第三方认证。所有成熟度评级仅来自公开披露。
[CR016, CR017, CR018, CR019, CR020, CR021]7.3 竞争压力、价格压缩与市场集中风险
Agilink 身处中国机器人生态里增速最快、也最拥挤的赛道。Humanoids Daily 报道称,中国目前有 150 多家人形机器人公司;中国国家发改委也在 2026 年直接警告行业存在「高度重复产品」以及「速度和泡沫」问题,说明北京可能从普惠式补贴转向选择性支持。政策一旦转向,后来者和单一产品公司承压最大。 竞争最尖锐的地方在价格。Agilink 创始人 Xiong Kun 将 OmniHand 2025 定价为 9,800 元人民币(约 $1,350 USD),有意压低中端价格带。但 Morgan Stanley 发现,92% 的中国企业买家要求单机成本低于 20 万元,才会大规模采用——这个门槛要由整套人形机器人系统跨过去,而不是单只灵巧手。也就是说,Agilink 的可服务市场同样被可负担性天花板卡住。与此同时,Unitree 于 2026 年 3 月在上交所科创板提交 IPO,拟募资 $580 million、即 42.02 亿元人民币,并披露 2025 年收入超过 17 亿元人民币(同比增长 335%)、净利润超过 6 亿元人民币。Unitree 已在规模下跑出盈利,这是 Agilink 必须对标的基准;其公开披露也会加速投资人审视其他中国机器人公司。 UBTECH 2025 年来自汽车和物流买方板块的订单簿超过 8 亿元人民币,证明全系统竞争者可以拿下 Agilink 同样瞄准的高价值制造客户。Goldman Sachs 2026 年对中国人形机器人厂商的调研指出,行业正从「通用想象」转向狭窄场景里的「专用落地」——安防巡逻、展厅导览、简单物流;通用型灵巧操作距离商业可行仍有 1–3 年。这意味着 Agilink 面向第三方 OEM 销售独立灵巧手的 TAM,可能比当前估值所暗示的时间点更晚兑现。 知识产权风险方面,灵巧手硬件一旦物料清单和电机 / 腱绳驱动架构披露,竞争者就能按成本复刻。Agibot 在 GitHub 托管 OmniHand-2025 SDK,会加速开源竞争侵蚀。Goldman Sachs 的硬件估值框架把数据飞轮(专有操作数据集)和软件栈视为主要长期护城河——Agilink 已把二者列为 2026 年路线图事项,但尚未商业化。[CR026, CR027, CR028, CR029, CR030, CR031]
该有向无环图追踪主要风险事件(立法、技术、合作伙伴)如何传导为收入受损和估值下行。
[CR001, CR003, CR016, CR026, CR034]7.4 运营、依赖与财务风险
Agilink 最结构性的集中风险,是它与母公司 Agibot 的关系。公开工商登记摘要确认,Agibot 在 Agilink 于 2026 年 1 月 14 日设立时持有约 80% 股权,Agibot 董事长兼 CEO Deng Taihua 担任法定代表人。四轮外部融资已稀释 Agibot 持股,但稀释后的确切所有权并未公开。影响很实质:Agilink 的研发人才、制造产能、数据采集基础设施(来自 Agibot 已部署的 10,000 多台机器人)以及早期收入渠道,都经由母公司流入。Agibot 若出现治理失灵、资本事件或战略转向,Agilink 的持续经营会直接受损。 投资人结构进一步放大了国家邻近型集中风险。Yicai Global 确认,Agibot 的投资人名单包括 BYD、SAIC Motor、BAIC Group、JD.com 和 Tencent。Agilink 自身的财团包括 SAIC Financial Holdings 和 Joyson Electronics(二者均为战略型上市产业投资人)。这种程度的国企敞口,会让任何拟议中的美国或欧洲二级交易面临类似 CFIUS 的审查风险,也意味着 Agilink 的商业决策可能受国企治理考量影响。 供应链风险又被 BIS 对先进计算 IC 的出口管制放大。OmniHand 内嵌推理芯片组的具体型号未公开,但 Agilink 的 AI 驱动触觉控制闭环需要推理硬件,可能落入 EAR 限制。如果中国国产半导体替代品达不到性能阈值,Agilink 的国际产品配置要么技术倒退,要么依赖许可证。 财务上,Agilink 估值已超过 $1 billion,而收入代理仅为 5,000 万至 2 亿元人民币(约 $7–27 million USD),对应的投机倍数说明投资人已把未来规模和软件可选性计入价格。Goldman Sachs 将年化生产成本下降 15–20% 视为机器人硬件经济模型跑通的门槛。在毛利率、烧钱速度或营运资本数据未披露前,剩余财务风险无法量化。IFR 指出,在重复性精密任务上,传统工业机器人仍比人形机器人更快、更可靠;这提高了一个风险:常规工业自动化中灵巧手的可服务市场,可能窄于整体机器人市场所暗示的空间。Crunchbase 将 Agilink 2026 年 5 月迈入独角兽的事件放在同月五家中国机器人独角兽的队列中,说明一旦行业情绪逆转,可能出现估值集群形成后的回撤风险。[CR034, CR035, CR036, CR037, CR038, CR039]
| 依赖项 | 交易对手 | 角色 | 集中度 | 失效场景 | 严重度 | 缓释 | 剩余暴露 |
|---|---|---|---|---|---|---|---|
| 母公司所有权和运营控制 | Agibot(设立时 80%) | 股东、法定代表人、研发管线、客户渠道 | 关键 | Agibot 治理失灵或战略转向,切断灵巧手供应和数据访问 | 关键 | 未披露 — 没有记录显示独立董事会或继任计划 | 关键 — Agilink 延续性与 Agibot 健康状况绑定 |
| AI 推理芯片供应 | 未披露芯片供应商 — 很可能受 BIS EAR 约束 | 触觉控制闭环的嵌入式计算 | 高 | BIS 将芯片加入 Entity List 或限制出口许可证 | 高 | 低 — 未披露芯片供应商多元化或国产替代规格 | 高 — 供应中断会让非中国市场 OmniHand 生产停摆 |
| 国资相邻战略投资者(SOE 集中) | 战略投资方:SAIC Financial Holdings、Joyson Electronics、BYD Ventures | 资本提供方、战略买家、供应链锚点 | 高 | SOE 投资者引导商业条款,偏离少数股东利益 | 中 | 部分 — 多个投资者降低单一 SOE 控制 | 中 — 若 SOE 阵营形成协调集团,治理风险上升 |
| 主要收入客户集中 | Longcheer Technology(唯一确认生产买家) | OmniHand 在消费电子制造中的终端客户 | 高 | Longcheer 以可靠性或成本为由暂停部署 | 高 | 无 — 未公开确认第二个生产客户 | 高 — 规模化前收入集中在单一客户 |
| 中国政府政策和补贴依赖 | MIIT、NDRC、地方政府补贴 | 研发补助、采购偏好、优惠市场准入 | 中 | NDRC 加快整合,并撤回对非头部玩家的补贴 | 中 | 部分 — Agibot/Agilink 的 MIIT 委员会席位带来政策可见度 | 中 — 选择性转向可能切断 Agilink 的政府采购管线 |
按严重度排序。集中度评级基于公开持股和投资者披露;Agibot 稀释后持股为估计值,因为后续轮次未披露。
[CR034, CR035, CR037, CR038, CR039, CR003]| 角色 / 职能 | 依赖或缺口 | 问题发生可能性 | 严重度 | 缓释 | 尽调路径 |
|---|---|---|---|---|---|
| CEO Xiong Kun — 唯一公开面孔和外部发言人 | 所有外部产品战略、客户推介和媒体关系都由 Xiong Kun 承担 | 中 | 高 | 未披露 — 没有联合创始人或 COO 的公开存在 | 核验继任计划、董事会构成和股权归属断崖期安排 |
| Agibot 人才管线和数据依赖 | 核心操控 AI 模型训练依赖 Agibot 车队部署数据;人才来自 Agibot 机器人团队 | 中 | 高 | 无 — 未披露数据共享协议或人才留存机制 | 索取数据许可条款和关键员工留任时间表 |
| Deng Taihua 担任法定代表人(Agibot 董事长 / CEO) | 设立时由 Agibot 关联法定代表人任职,削弱治理独立性 | 低 | 高 | 无 — 未公开披露公司治理改革 | 核验剥离后法定代表人是否已转给 Xiong Kun 或独立董事 |
| MIIT 标准利益冲突(Peng Zhihui) | Agibot 联合创始人在 MIIT 委员会任职,可能造成标准机构俘获或声誉冲突 | 低 | 中 | 部分 — 委员会有多方参与(65 名成员,Unitree 也有代表) | 监测已发布 MIIT 标准草案,寻找可能有利于 Agibot 供应链的接口规格 |
仅基于公开披露。角色描述反映公开报道;内部治理、继任和期权池细节未披露。
[CR034, CR035, CR011, CR036]这张有向图梳理 Agilink 的关键依赖:所有权、资本、R&D 和收入关系。一旦任何一环断裂, 运营都会受损。
[CR034, CR037, CR038]7.5 缓释措施、监测指标与终止标准
Agilink 有几项结构性缓释因素,能部分抵消风险堆叠。MIIT 委员会参与(Peng Zhihui 担任副主席)让 Agibot 生态公司能提前看到标准制定方向,降低中国侧合规的监管意外。四轮融资节奏——包括来自汽车、电子和互联网行业的战略投资人——提供了多条资本路径,即便单一投资人退出也不至于断档。Longcheer Technology 已确认 99.9% 成功率,虽然只限一个现场,但给投资人提供了可引用的近期商业可信度样本。 但地缘政治风险的缓释成熟度很低。已审阅来源没有披露任何针对 ASRA 通过、EU AI Act 合规工程,或内嵌芯片 BIS 许可证采购的应急方案。面向海外企业客户的中国 NI Law 数据主权问题,也没有披露架构回应。这些都是实质性尽调事项。 投资论点恶化的可监测触发因素包括:(1)ASRA 走出委员会或签署成法——这会关闭美国联邦采购,并很可能引发政府以外企业买家的寒蝉效应;(2)Unitree IPO 形成行业市值基准,暗示 Agilink 的收入倍数应更低;(3)任何披露的灵巧手生产环境产品安全事故;(4)Agibot 科创板 IPO 文件披露 Agilink 持股低于 40%,意味着稀释可能让母公司激励错位;(5)BIS 将 Agilink 或其芯片供应商加入 Entity List,或发布更严格的 IC 规则。 投资论点的终止标准是:Agilink 失去主要渠道(Agibot 崩溃或切断灵巧手供应协议);任何 G7 司法辖区出现重大产品责任判决,并确立自主操作中制造商责任先例;或中国政府从人形机器人部件撤回补贴,触发单位经济模型反转。[CR010, CR011, CR019, CR021, CR029, CR037]
| 风险 | 可监测触发因素 | 阈值 / 事件 | 行动含义 |
|---|---|---|---|
| ASRA 地缘政治升级 | 参议院 / 众议院关于中国机器人法案的投票日程,或行政令 | ASRA 通过委员会,或被签署成法 | 退出面向美国的头寸;在假设美国联邦收入为零下重估 Agilink TAM |
| Unitree IPO 估值压缩 | Unitree STAR Market 上市价和 IPO 后 P/S 倍数 | Unitree IPO 暗示行业 P/S 低于 5x;或 Unitree 股票上市前 90 天跌幅超过 30% | 重新审视 Agilink 估值模型;考虑退出区间低于 $500M 的下行情景 |
| 产品安全事件 | 涉及灵巧机械手的 G7 工作场所安全报告、召回或产品责任索赔 | 任何公开报道的中国灵巧手产品伤害事件或监管停发令 | 暂停投资;启动产品责任深度尽调和安全架构审查 |
| Agibot 母公司不稳定 | Agibot STAR Market IPO 招股书披露 Agilink 持股低于 40%;或 Agibot 资本事件(降估值融资、监管暂停) | Agibot 披露的 Agilink 持股降至 40% 以下,或 Agibot STAR IPO 被撤回 | 重新评估 Agilink 运营独立性;建模 Agibot 渠道流失情景 |
| 出口管制升级 | BIS Entity List 新增名单,或影响 Agilink 嵌入式芯片组供应商的新 IC 规则 | Agilink 或主要芯片供应商被加入 Entity List;或 BIS IC 规则收紧,取消授权设计方豁免 | 启动芯片供应商审计;评估中国国产半导体替代可行性 |
触发定义基于可公开监测的信号。行动含义面向财务投资者;战略投资者可能有不同阈值。
[CR001, CR003, CR028, CR029, CR034, CR039]7.6 附录
08估值
8.1 Agilink 的 $1 Billion 独角兽标记:背景与证据质量
Agilink 于 2026 年 5 月跨过独角兽门槛;自 2026-01-14 设立以来完成第四轮融资后,据报估值超过 $1 billion。唯一直接确认 $1 billion 数字的英文一手来源,是 Crunchbase News 2026 年 6 月 1 日文章,文中写道:「上海的 Agilink 自 1 月从 Agibot 拆分以来已完成四轮融资,并达到 $1 billion 估值。」该文没有说明轮次规模、pre-money 还是 post-money 口径,也没有列出第四轮投资人。多家中文商业媒体(行业媒体、Pedaily 类媒体)在 2026 年 5 月 19–20 日窗口期报道了同一估值跃升,印证 Crunchbase 说法,但没有任何一家发布经审计数据。 这种证据画像——投资人报道的独角兽,却无收入披露——在 2025–2026 年机器人融资潮中很常见。当时全球投入 $14 billion,仅中国就投入 $5.6 billion,但对当期经济模型的基本面承销很有限。Agilink 四轮融资速度(五个月成为独角兽)在灵巧手部件子赛道没有直接先例,不过与机器人手臂类比中的 Tianji 时间线相比更有利。投资人阵容——Hillhouse Venture、BlueRun、Baidu Ventures、Yunfeng Capital、SAIC Financial Holdings、Joyson Electronics 和 C Capital——按中国 VC 标准质量很高,更符合抢份额优先的竞赛姿态,而不是回报纪律优先。由于没有公开说明 pre/post-money 区分,任何进入价格分析都必须把 $1 billion 视为方向性标记,并根据轮次规模留出 ±20% 的模糊带。[CV001, CV002, CV003, CV004, CV005, CV016]
8.2 可比估值组:全系统、部件层与中国市场
为 Agilink 构建相关可比组,必须拆成三层:西方全系统人形机器人、中国全系统人形机器人,以及中国部件层供应商。每一层的收入画像、战略逻辑和市场折价都不同。 西方全系统组(Figure AI、Apptronik、Agility Robotics、1X Technologies)估值分散很大,从 Agility 的 $1.75 billion 到 Figure AI 的 $39 billion。Figure AI 2025 年 9 月 Series C——由 Parkway Venture Capital 领投,Sacra 档案确认 post-money 估值超过 $39 billion——是单一投资人标记驱动的异常值,二级市场参与有限,不能作为可靠的行业锚。Apptronik 2026 年 2 月的 $5 billion 估值(公司官方新闻稿;SEC Form D 印证了相关 Series A Special Purpose Vehicle)由更广泛财团支持,战略投资人包括 Google DeepMind、Mercedes-Benz、John Deere 和 Qatar Investment Authority。Agility Robotics 在 Series C 后约 $1.75 billion post-money,按总融资规模看是最接近 Agilink 的结构性可比,但其部署记录(GXO/Amazon Fulfillment)和全系统产品范围,使它只能部分类比。 中国市场里,Unitree 是唯一有公开申报收入数据的公司:2025 年收入超过 17 亿元人民币($235 million),同比增长 335%,净利润超过 6 亿元人民币($83 million),支撑其拟募资 $580 million 的上交所科创板 IPO。Unitree 隐含的 IPO 阶段倍数(按募资额相对于科创板先例估算,为 2025 年收入的 8–12x)为行业倍数提供了最扎实的下限。Agibot(Agilink 母公司)在 2025 年 8 月 LG/Mirae Asset 轮按 Pandaily 报道 pre-money 估值超过人民币 70 亿元(约 $960 million),Tracxn 则引用了更近期的 $6.4 billion 估计。部件层里,Tianji 2026 年 5 月的 $1.5 billion Series B,是 Agilink 自身融资轮最相关的结构性可比。[CV006, CV007, CV008, CV009, CV010, CV011]
| 公司 | 估值(最新) | 阶段 / 轮次 | 已披露收入 | 业务类型 | 与 Agilink 的相关性 | 关键限制 |
|---|---|---|---|---|---|---|
| Figure AI(美国) | $39B(2025 年 9 月) | C 轮 | None | 全系统人形机器人,美国 | 西方可比公司的上沿;AI 集成溢价 | 单一投资人异常值(Parkway VC);融资额相对估值仅 1x,极端偏高 |
| Apptronik(美国) | $5B(2026 年 2 月) | A 轮延展 | None | 全系统人形机器人,美国 | 西方可比公司,财团质量较好;可作战略投资人质量基准 | 累计融资 $935M;没有商业收入;NASA 背景带来溢价 |
| Agility Robotics(美国) | ~$1.75B(2025 年 3 月) | C 轮 | None | 全系统人形机器人,美国 | 按累计融资额看,定价最接近的西方可比公司 | Amazon 支持落地,但全系统范围不同于 Agilink 组件定位 |
| 1X Technologies(挪威) | ~$0.5–1B 估算 | B 轮 | None | 全系统人形机器人,全球 | OpenAI 支持;全球市场溢价 | 估值为推算;仅融资 $100M;家用 / 商业安防细分市场 |
| Unitree(中国) | ~$3–5B 估算(2026) | Pre-IPO | $235M(2025) | 全系统人形机器人,中国 | 唯一披露收入的中国可比公司;IPO 锚定公开市场底部 | 全系统硬件,不是组件供应商;STAR Market 语境 |
| Agibot(母公司,中国) | $6.4B 估算(2026) | E 轮等效 | None | 全系统人形机器人,中国 | 直接母公司;生态重叠;持股 80% | 母公司;存在并表风险;估值包含 Agilink 子公司 |
| Tianji(中国) | $1.5B(2026 年 5 月) | B 轮 | None | 机器人手臂 / 组件,中国 | 结构上最接近的中国组件层可比公司(Series B,2026) | 做机器人手臂而非灵巧手;执行器产品范围更宽 |
| Agilink(本报告) | >$1B(2026 年 5 月) | 第 4 轮 | None | 灵巧手,中国 | 标的公司 | 未说明投前 / 投后;无收入;仅有单一英文来源 |
估值来自引用来源披露或分析师估算,均未获审计财务验证。中国公司估值反映私募轮次标记,不是 STAR Market 定价。'est.' = 估算;除非有一手来源另行确认。
[CV006, CV007, CV008, CV009, CV010, CV011]对比可比公司组中披露或估算的估值,区间从 $0.1B 到 $39B;Agilink 处在这条谱系的低端。
估值来自投资者标记或分析师估计,尚未独立验证。Figure AI $39B 是 2025 年 9 月 Series C 的 post-money 估值。Unitree 取 pre-IPO 估值区间中点。Agibot $6.4B 来自 Tracxn 估计。 1X Technologies $0.75B 由分析师根据 $100M 融资反推。所有数字均为十亿美元。
[CV006, CV009, CV011, CV012, CV013, CV015]8.3 估值框架:硬件、平台与战略期权方法
Agilink 这个阶段的硬件公司,可用四个框架估值:收入倍数(P/S)、单位出货代理价值、平台 / 软件期权价值,以及收购方战略价值。收入未披露时,没有一个框架能给出稳定锚点。 传统硬件 P/S 倍数需要知道收入。按出货代理估算——约 8,000 只灵巧手,混合 ASP 为 9,800–30,000 元人民币,意味着累计收入约 7,800 万至 2.4 亿元人民币($10–33 million USD)。在 $1 billion 估值标记下,隐含收入倍数为累计收款的 37–100x,远高于 Series C 阶段硬件创业公司常见的 5–15x 前瞻收入区间。这个差距并不自动构成否定:Goldman Sachs 指出,2025 年以来私人机器人估值主要由数据飞轮、制造规模和部署证明驱动,而非当期经济模型。IDC 预测到 2030 年人形机器人出货量超过 510,000 台、CAGR 95%,为前瞻 TAM 论证提供了分母。 平台 / 软件期权价值视角更宽容。如果 Agilink 成功把专有操作数据集和 AI 模型授权变现给第三方 OEM——按 Xiong Kun 的说法,这是已陈述但尚未商业化的路线图事项——适用倍数会从硬件(5–15x)转向 AI 平台(20–50x 前瞻收入)。Morgan Stanley 的部件供应商分析特别指出 Inovance 和 Leaderdrive 是机器人供应链最早受益于盈利的公司,验证了部件层论点。对收购方战略价值而言,拥有专有灵巧手栈、8,000 多个在场单元和 3D 触觉感知的护城河是真实存在的,但没有披露 IP 登记和毛利率,就无法量化。[CV019, CV020, CV021, CV022, CV023, CV024]
| 框架 | 最适用阶段 / 公司类型 | 主要驱动因素 | Agilink 代理值 | 适用性评级 |
|---|---|---|---|---|
| 收入倍数(P/S) | 已披露收入的硬件初创公司 | 远期收入 5–15x | ~37–100x 隐含累计倍数(投机性) | 低——未披露收入 |
| 单位出货代理值 | 早期、预收入硬件公司 | ASP × 销量 × 估算利润率 | $1B / 8K 台 = 每台已交付产品隐含 $125K | 中——仅作方向判断;出货量未经审计 |
| 平台 / AI 期权价值 | AI 软件和基础模型公司 | 生态杠杆;可防守的数据集和模型护城河 | 授权平台变现后可达 20–50x | 高度理想化;尚未成为商业产品 |
| 可比私募轮次基准 | 同阶段 VC 市场基准 | 同赛道、同地域、同融资规模的近期可比公司 | $1B 与中国组件层 B–C 轮区间相符(Tianji:$1.5B) | 中——在无收入数据下,这是最可用的锚 |
| 收购方战略价值 | M&A 场景;技术收购 | 对 NVIDIA、Tesla、Agibot 或 OEM 等效收购方而言,自有灵巧操作栈的价值 | 难以估算;战略溢价真实存在但无法量化 | 牛市情形下为高;没有披露 IP 目录则无法核验 |
| 相对全系统人形机器人的折价 | 组件供应商对比全系统 | 全系统集成商享有整机溢价;灵巧手约为 6 个主要子系统之一 | 按全系统可比公司 10–15% 计,相对 Agility/Apptronik 隐含 $0.3–1.2B 公允区间 | 中——结构性可用;受 Agilink 跨 OEM 潜力限制 |
各框架并不互斥。本轮周期里的投资人标记混合了平台期权价值和可比基准,而不是收入倍数。表格只是比较视角,不是混合估值模型。
[CV019, CV020, CV025, CV026, CV030, CV031]基于不同商业化假设,展示 Agilink 到 2028 年的低位、基准和高位估值情景。
区间估算方法是把可比公司倍数套用到不同情景下的收入预测,并非 DCF 模型。入场估值按 $1B; 相对入场价的回报为扣除稀释和费用前的毛回报。所有数字均为十亿美元。
[CV039, CV040, CV041]8.4 情景分析:牛市、基准与熊市估值路径
到 2028 年,三种情景圈定了 Agilink 可支撑股权价值的范围;关键变量是授权平台多快兑现,以及人形机器人市场能否维持当前投资人情绪。 牛市情景假设 Agilink 向 10 家或更多外部 OEM 客户授权,到 2027 年年交付量超过 100,000 单元,并把其操作数据集转化为面向全球机器人制造商的经常性 AI 模型订阅。在这些假设下,每年 $150–300 million 收入、35–45% 毛利率,可按 20–30x 前瞻收入倍数支撑 $3–8 billion 估值,与 Apptronik 的溢价以及 Goldman Sachs 对中国机器人龙头的平台价值框架一致。中国政策顺风会放大这一情景——IFR 确认,中国第十五个五年规划把机器人置于国家战略核心,截至 2024 年国内供应商份额为 57%。 基准情景假设当前 $1 billion 标记在 2026–2027 年大致稳定:公司推进硬件规模化,从 Agibot 供应渠道和两到三家外部客户获得收入,但因 AI 模型成熟度缺口推迟授权平台。按每年 20,000–50,000 单元计算,基于出货的收入为 $27–135 million USD(混合 ASP 9,800 元人民币,折算美元),在没有重大竞争者冲击的前提下,可用传统硬件倍数支撑 $800 million 至 $1.5 billion 估值。这个基准情景与 Unitree 2025 年收入轨迹作为行业地板相一致。 熊市情景出现在几种情况下:机器人估值泡沫消退,Agibot 决定重新整合灵巧手供应链,或 sim-to-real 准确率缺口拖慢商业化——IDC 报告显示,准确率从仿真中的 80–90% 跌到真实部署中的低于 50%。下行周期里,部件供应商的倍数压缩最快,因为全系统 OEM 会把部件内制以收回利润。在这一情景中,NDRC 对「高度重复产品」和 150 多家竞争者的警告会抹掉溢价,把估值压缩到 $100–400 million 的困境硬件倍数。[CV028, CV034, CV035, CV038, CV039, CV040]
| 论点 | 支撑证据 | 何种情况会改变判断 |
|---|---|---|
| 灵巧手瓶颈释放 TAM | 约 80% 高价值操控任务需要尚未标准化的灵巧度(MIIT/Agibot) | 仿真到真实差距在没有专有触觉传感器优势下收敛 |
| 最快实现万元人民币以下规模化生产 | 8,000+ 只手;OmniHand 2025 售价 9,800 RMB — 比上一代便宜 7–150x | Unitree 或其他竞争者在 12 个月内以低于 5,000 RMB 进入 |
| Agibot 管线保障近期收入 | Agibot 全球出货第一(Omdia);Agilink 以转移定价向母公司供应灵巧手 | 关联方收入按公允距离交易定价被侵蚀;利润率无法验证 |
| 战略投资者基础降低退出路径风险 | Hillhouse、Yunfeng、Baidu Ventures、SAIC、Joyson;汽车和电子 OEM 参与 | 投资者基础过度集中在中国;全球机构后续跟投有限 |
| 未披露收入 — 倍数具有投机性 | 无 CSRC 备案;无审计账目;$1B 估值只有单一 Crunchbase 英文来源 | Agilink 披露管理账目或向监管机构申报 |
| 泡沫退潮中的组件价格压缩风险 | NDRC 2026:“速度与泡沫”;150+ 竞争者;Goldman Sachs 2026 将预期重置到近期现实 | Unitree IPO 证明公开市场胃口;人形机器人 TAM 比熊市情景更快兑现 |
第 1–4 行是支持投资的论点;第 5–6 行是按风险加权的反方论点。各行不按概率排序。
[CV019, CV023, CV024, CV034, CV040, CV041]| 情景 | 概率信号 | 关键假设 | 估值区间(2028) | 主要风险 |
|---|---|---|---|---|
| 牛市 | 低–中 | 10+ 家 OEM 授权客户;到 2027 年年出货 100K+ 台;AI 模型授权形成经常性收入;毛利率落到 35–45% | $3–8B | 执行速度;国产仿制者侵蚀 IP 护城河 |
| 基准 | 中 | 年出货 20–50K 台;Agibot 渠道贡献 >40% 收入;授权延后 12–18 个月;2 个外部客户背书 | $0.8–1.5B | 板块情绪降温后估值倍数压缩;母公司渠道集中 |
| 熊市 | 低–中 | 泡沫退潮;Agibot 重新整合;仿真到现实的落差卡住商业化;NDRC 提高选择性导致补贴减少 | $0.1–0.4B | 资本约束逼出降价轮;全系统 OEM 挤压组件层利润率 |
概率信号只指方向,并未校准。区间按可比倍数估算,不构成正式 DCF。所有估值都假设标的公司没有披露一手收入。
[CV039, CV040, CV041, CV042]按当前估值,从七个投资维度给 Agilink 打分,可直接提交 IC 讨论。
分数是作者基于本章和前文 01–07 章证据作出的判断。量表为 1–10,10 代表最强。分数反映 截至 2026-06-16 可获得的证据;若出现重大披露,应更新。
[CV001, CV022, CV023, CV026, CV034, CV038]8.5 战略溢价与折价因素:中国、治理与部件层
Agilink 估值受四类系统性溢价 / 折价因素影响;这些因素不能简单相互抵消,必须合并权衡。 第一,相比同等西方市场估值,中国来源资产要承受地缘政治折价。American Security Robotics Act(两党参议院法案,2026 年 3 月)明确瞄准中国人形机器人厂商,BIS 对先进计算 IC 的出口管制也会给任何内嵌 AI 推理处理器的部件制造供应链摩擦。Unitree 的 pre-IPO 讨论显示,在基本面相当时,中国 A 股科创板机器人公司相较美国 VC 标记折价 30–50%。在 Agilink 这个阶段,中国第十五个五年规划匹配度和国内供应链成本优势会部分抵消这项折价。 第二,应给出关联方治理折价。Agibot 在 Agilink 设立时保留约 80% 股权,Deng Taihua(Agibot 董事长)担任法定代表人。缺少独立董事会、披露的优先权结构或少数股东保护契约,意味着外部少数股东的治理救济有限。这一风险类似早期 Unitree 和 UBTECH 的治理画像——二者在科创板上市前都需要完成治理整改。 第三,相比全系统人形机器人估值,部件层结构性折价适用。以 Agilink 当前硬件优先阶段看,适用市场是完整人形机器人系统价值的 10–15%(灵巧手是五六个主要子系统之一)。只有当 Agilink 建立跨 OEM 供货关系,且任何单一全系统玩家都无法内部复制时,末端执行器专精带来的战略溢价才会释放。 第四,先发溢价真实存在,但很脆弱。没有其他中国灵巧手厂商能以低于 10,000 元人民币的价格,在这种形态下出货 8,000 多只带触觉反馈的单元。Unitree 数字争议(5,500 台 vs 5,168 台,未验证)说明,全行业口径都未经审计;Agilink 的数字也有同样提示。如果溢价真实,就必须靠 IP 申请、制造护城河和专有训练数据守住——这些从外部都无法公开验证。[CV028, CV029, CV030, CV031, CV032, CV035]
从现有证据出发,串起投资假设、风险评估,最后落到估值偏高下的“继续研究”建议。
逻辑链做了简化;实际决策还包含这里无法完整呈现的定性因素。
[CV001, CV005, CV028, CV029]8.6 投资建议、终止标准与最终尽调要求
建议为继续研究,置信度中等,风险评级高,估值立场偏紧。偏紧不等于绝对意义上的高估——投资人基础、融资速度和灵巧手 TAM 论点都可信——而是说,当前公开证据不足以让人有把握地承销 $1 billion 进入价格。「偏紧」和「合理」的区别取决于三项披露:经审计收入、包含 pre/post-money 说明的股权结构明细,以及至少一个外部(非 Agibot)客户证明。 若出现以下情况,建议可上调为买入:(a)Agilink 向 CSRC 提交财务文件,或以其他方式披露收入,且增速与出货代理一致;(b)后一轮融资由新的机构投资人按高于 $1.5 billion 定价,确认估值标记;或(c)Agibot 科创板 IPO S-1 披露单独验证 Agilink 收入贡献。若出现以下情况,应下调为回避:(a)Agibot 重新整合灵巧手供应链,或以低于 $1 billion 条件收购 Agilink;(b)BIS 对 OmniHand 使用的内嵌推理芯片组实施出口管制;或(c)Agilink 下一轮融资为 down-round。终止标准表记录了七个带具体阈值的触发因素。尽调要求表列出按当前估值投入资本前必须取得的六项最低证据。[CV001, CV004, CV027, CV038, CV039, CV040]
| 触发器 | 阈值 / 事件 | 对投资论点的传导 | 行动含义 |
|---|---|---|---|
| Agibot 重新整合供应链 | 以低于 $1B 的条件战略收购 Agilink 剩余股权,或宣布内部制造 OmniHand 等效产品 | 独立股权价值坍塌;少数股东回报路径被切断 | 立即退出,或硬性拒绝新增资本投入 |
| 行业倍数压缩 | Unitree SSE IPO 表现低于 2025 年收入 10x,或 Figure AI 二级市场跌破 $10B | 投资人给预收入机器人公司的溢价系统性退潮 | 持有;降低敞口;重定价入场纪律 |
| 关键人物离职 | Xiong Kun 离开,90 天内没有指定继任者 | 技术路线图可信度受损;LP 沟通风险上升 | 大幅下调信心;重新尽调继任安排 |
| BIS 对嵌入式芯片实施出口管制 | OmniHand 推理处理器被确认归入 BIS 适用分类 | 国际市场实质关闭;只剩中国单一市场论点 | 降低总仓位;按中国单一市场估值底部重新评估 |
| E 轮前仍未披露收入 | 到 2026 日历年结束仍无公开财务 KPI | 估值无法核验;重复信息不足下入场的风险 | 维持 research-more;披露前暂缓承诺 |
| 降价轮 | 下一轮定价由任何机构投资人给出,投后估值低于 $1B | 成熟投资人信心信号转弱;前轮标记受损 | 进入退出观察;立即重新评估 |
触发器大致按影响的即时性排序。阈值为指示性口径,应按实际投资协议和报告要求校准。
[CV028, CV029, CV038, CV041]| 议题 | 缺失证据 | 重要性 | 责任人 / 尽调路径 |
|---|---|---|---|
| 收入和毛利率 | 任何期间都未披露 ASP 组合、总收入、毛利率或烧钱速度 | 没有可核验收入基底,无法承销 $1B 标记;37–100x 倍数带有投机性 | 向领投方(Hillhouse)索取管理账;调取 CSRC 登记中的任何自愿披露 |
| 股权结构和轮次条款 | 四轮融资的投前估值与投后估值;清算优先级;反稀释条款;员工期权池 | 无法定价入场稀释,也无法评估降价轮场景下的保护 | 直接询问 CFO / 法律顾问;查询 CSRC 国家企业信用信息公示系统 |
| 客户集中度(关联方与外部) | 未拆分来自 Agibot(关联方)与独立客户的收入 | 决定护城河是真实存在,还是依赖持股 80% 的母公司渠道 | 访谈非 Agibot 客户;要求分部报告 |
| IP 归属和 Agibot 授权条款 | 手部相关专利和设计权;从 Agibot 承继 IP 的授权条款 | Agibot 持股 80% 不等于独占 IP;授权终止是结构性风险 | 专利律师审查 CNIPA 申请;合同尽调 IP 转让与授权 |
| 相对 Agibot 的竞争定价(关联方转移定价) | 出售给 Agibot 的灵巧手按市场价,还是低于市场的转移价格? | 转移定价操纵会人为抬高 Agilink 表观单位经济性 | CFO 访谈;未来任何申报文件中的关联方披露;对标 Tianji 定价 |
| 可比运营指标基准 | Unitree IPO 招股书(SSE 披露,2026)包含唯一公开可得的中国人形机器人运营指标 | 为工业规模下单位经济性的真实水平提供最扎实的底部 | 获取并分析 Unitree SSE 招股书(sse.com.cn);与 IPO 分析师交叉核验 |
尽调清单按对当前估值的重要性排序。第 1 和第 2 项会阻断当前阶段的任何资本承诺。
[CV004, CV005, CV027, CV029]8.7 附录
免责声明
本报告是基于公开证据的尽调快照,不构成投资建议。重要的财务、法律、技术和合同事实仍未公开; 作出任何投资决定前,应直接向管理层和一手文件核验。
证据索引
| 编号 | 陈述 | 可信度 | 来源 |
|---|---|---|---|
| CO001 | Agilink's independent legal entity was incorporated in Shanghai on 2026-01-14. | 中 | SO001, SO004 |
| CO002 | Reviewed public sources identify the company as Shanghai Linjiedian Innovation Intelligent Technology Co., Ltd. | 中 | SO001, SO004 |
| CO003 | Agilink is based in Shanghai, China. | 中 | SO001, SO014 |
| CO004 | Agilink was spun out from Agibot's dexterous-hand business as an independent entity in January 2026. | 中 | SO002, SO004, SO009 |
| CO005 | Agilink focuses on dexterous robotic hands and related end-effectors rather than full humanoid robot bodies. | 中 | SO002, SO010, SO016 |
| CO006 | Pedaily reported that Agibot contributed RMB 4 million and held 80% of Agilink at incorporation. | 中 | SO001 |
| CO007 | Public registry summaries cited by Sina Finance say Agilink was formed through Agibot-affiliated entities rather than as a fully orphaned spinout. | 中 | SO004 |
| CO008 | Xiong Kun said the spinout structure was meant to support continued R&D spending and wider external sales. | 中 | SO002, SO010, SO011 |
| CO009 | Agilink publicly positions its products for embodied-AI, industrial automation, logistics, service-robotics, and research scenarios. | 中 | SO002, SO009, SO016 |
| CO010 | No reviewed public source disclosed Agilink's revenue split, pricing architecture at company level, or recurring-services mix. | 中 | SO001, SO002, SO003, SO004 |
| CO011 | Xiong Kun is the principal public operating leader associated with Agilink in reviewed February 2026 reporting. | 中 | SO002, SO010, SO011 |
| CO012 | Xiong Kun joined Agibot in November 2024 to run the dexterous-hand business before the spinout. | 中 | SO002, SO010 |
| CO013 | Reviewed February 2026 reporting says Xiong Kun previously worked at Tencent Robotics X. | 中 | SO002, SO010 |
| CO014 | Reviewed February 2026 reporting says Xiong Kun also held roles at IDEA Research Institute and Inovance before joining Agibot. | 中 | SO002, SO010 |
| CO015 | Deng Taihua was named as Agilink's legal representative in public registry summaries cited by January and May 2026 media coverage. | 中 | SO001, SO004 |
| CO016 | No reviewed source disclosed an independent Agilink board, outside directors, or founder-level succession plan. | 中 | SO001, SO002, SO004, SO009 |
| CO017 | Agilink's governance profile remained parent-linked at launch because operational independence was disclosed without matching board or cap-table detail. | 中 | SO001, SO004, SO006 |
| CO018 | Key-person dependence is material because Xiong Kun is the main named operator while public disclosures on broader management remain thin. | 中 | SO002, SO010, SO011 |
| CO019 | Pedaily reported on 2026-01-26 that Agilink had already completed two financings shortly after incorporation. | 中 | SO001 |
| CO020 | Pedaily reported that the January 2026 financings were jointly led by Hillhouse Venture and BlueRun Ventures. | 中 | SO001 |
| CO021 | Agilink said after the January 2026 financings that it was planning additional strategic financing. | 中 | SO001 |
| CO022 | Multiple outlets reported that Agilink completed another several-hundred-million-yuan round led by a major internet company in February 2026. | 中 | SO002, SO005, SO010, SO011 |
| CO023 | The February 2026 investor group included Baidu Ventures, Yunfeng Capital, Synstellation Capital, Joyson Electronics, Longcheer Technology, and SAIC Financial Holdings. | 中 | SO002, SO005, SO010, SO011 |
| CO024 | By 2026-05-19 and 2026-05-20, reviewed media coverage converged that Agilink's valuation had surpassed US$1 billion. | 中 | SO003, SO004, SO007, SO008, SO009 |
| CO025 | May 2026 reporting said Agilink had completed four financing rounds since its January 2026 incorporation. | 中 | SO003, SO004, SO006, SO007 |
| CO026 | The latest disclosed funding proceeds were earmarked for dexterous-hand large-model R&D, open datasets, and continued hardware iteration. | 中 | SO003, SO007 |
| CO027 | Reviewed public reporting did not disclose Agilink's total lifetime capital raised, debt facilities, or secondary-share transactions. | 中 | SO001, SO002, SO003, SO004, SO007 |
| CO028 | Agilink became a unicorn less than five months after incorporation. | 中 | SO003, SO004, SO007 |
| CO029 | By May 2026, multiple outlets reported cumulative deliveries of more than 8,000 dexterous hands. | 中 | SO007, SO008, SO009, SO016 |
| CO030 | The same May 2026 coverage reported cumulative deliveries of more than 10,000 grippers. | 中 | SO008, SO009, SO016 |
| CO031 | Company and channel materials identify 2025 as the year OmniHand and OmniHand Pro entered the market. | 中 | SO002, SO018, SO019, SO022, SO024 |
| CO032 | The OmniHand 2025 base model has 10 active and 6 passive degrees of freedom and weighs at most 500 grams according to official product and SDK materials. | 中 | SO018, SO021 |
| CO033 | Official and distributor materials describe OmniHand 2025 as supporting CANFD communication, ROS 2 integration, and 400-plus tactile sensing points. | 中 | SO018, SO021, SO023 |
| CO034 | OmniHand Pro 2025 is positioned as a higher-end industrial hand with 12 active and 19 total degrees of freedom. | 中 | SO019, SO020, SO024 |
| CO035 | Agilink's product stack spans multi-DoF five-finger hands and more cost-effective grippers across several performance tiers. | 中 | SO002, SO011, SO016 |
| CO036 | APC 2026 introduced OmniHand 3 Ultra-T as a 22+3 DoF tendon-driven flagship with full-hand 3D tactile sensing, an integrated palm camera, sub-0.3 second response time, and a 10:1 load-to-weight ratio. | 中 | SO015, SO025 |
| CO037 | No reviewed public source disclosed Agilink's revenue, ARR, or gross margin. | 中 | SO003, SO004, SO007, SO009 |
| CO038 | No reviewed public source disclosed Agilink's headcount or customer count. | 中 | SO003, SO004, SO007, SO009 |
| CO039 | Public evidence is strong on fundraising momentum and shipment volume but weak on cap-table mechanics and unit economics. | 中 | SO003, SO004, SO007, SO009 |
| CO040 | An adverse June 2026 sector source said Chinese robotics planners were warning against bubbles and repetitive products in the humanoid ecosystem, a backdrop that raises execution pressure on Agilink. | 中 | SO027 |
| CM001 | Agilink's primary market is the dexterous robotic hand and intelligent end-effector segment, encompassing multi-fingered manipulation modules with embedded AI, tactile/force sensing, and proprietary control software for humanoid and mobile-manipulation platforms. | 中 | SM022, SM006 |
| CM002 | The dexterous hand market excludes traditional single-axis pneumatic grippers without embedded AI, fixed-arm industrial robots optimized for repetitive tasks, and human manual inspection services. | 中 | SM010, SM005 |
| CM003 | Humanoid robot platform adoption serves as the primary demand pull for dexterous end-effectors; end-effector revenue is embedded within and dependent on humanoid hardware spend, with the hand estimated at approximately 10–15% of total platform cost (unvalidated proxy assumption). | 低 | SM004, SM006 |
| CM004 | Goldman Sachs Research estimates the global humanoid robot hardware market may reach at least $6 billion in 10–15 years (by ~2035E) in a base case, filling 4% of the US manufacturing labor shortage gap by 2030E and 2% of global elderly-care demand by 2035E. | 高 | SM001, SM002 |
| CM005 | Goldman Sachs estimates a mid-scenario global humanoid robot hardware market of approximately $38 billion by 2035E. | 高 | SM001, SM009 |
| CM006 | Goldman Sachs' blue-sky scenario estimates global humanoid hardware revenue at up to $154 billion by 2035E—close to the global EV market size—conditional on fully overcoming product design, affordability, technology, and public-acceptance hurdles. | 高 | SM001, SM009 |
| CM007 | MarketsandMarkets estimates the global humanoid robot market at $2.92 billion in 2025, growing to $15.26 billion by 2030 at a 39.2% CAGR. | 中 | SM003, SM004 |
| CM008 | MarketsandMarkets estimates the China humanoid robot sub-market at $0.40 billion in 2025, growing to $2.80 billion by 2030 at a 47.6% CAGR—the fastest-growing national segment in their model. | 中 | SM003, SM005 |
| CM009 | IDC forecasts global humanoid robot shipments will exceed 510,000 units by 2030, representing a compound annual growth rate of approximately 95% from the 18,000 units shipped globally in 2025. | 高 | SM004, SM013 |
| CM010 | IDC reports approximately 18,000 humanoid robots were shipped globally in 2025, with more than 85% of deployments concentrated in performances, education, data collection, and guided-tour service scenarios rather than industrial production. | 高 | SM004, SM020 |
| CM011 | Chinese vendors accounted for approximately 95% of global humanoid robot shipments in 2025, establishing a dominant manufacturing position and near-monopoly on early supply. | 高 | SM004, SM005 |
| CM012 | IDC identifies dexterous hands—critical for fine manipulation—as among the fastest-growing hardware categories within humanoid robotics, explicitly "poised for rapid development" in the 2026 forecast horizon. | 高 | SM004, SM006 |
| CM013 | Goldman Sachs ($6B base/$38B mid/$154B blue-sky by 2035) and MarketsandMarkets ($15.26B by 2030) provide materially different market size estimates; the divergence reflects incompatible scope definitions—hardware-only revenue vs. broader market including software—and different time horizons. | 中 | SM001, SM003 |
| CM014 | Goldman Sachs estimates $38 billion in humanoid robot hardware revenue by 2035, while ARK Invest estimates up to $24 trillion in economic value from humanoid labor displacement; the 600x divergence reflects incompatible metrics—hardware-sale revenue versus labor-displacement GDP value—and the two figures must not be compared or summed. | 中 | SM001, SM009 |
| CM015 | Agibot co-founder and MIIT Humanoid Robot Standards Technical Committee deputy director Peng Zhihui stated that "nearly 80 percent of tasks where humans excel but traditional automation struggles are strongly related to tactile sensing"—directly identifying the unmet demand that Agilink's dexterous hand addresses. | 中 | SM006, SM015 |
| CM016 | China's 15th Five-Year Plan (2026–2030) places robotics at the heart of its modern industrial system, marking the first time robotics is elevated to primary framework-level priority; all subordinate sectoral and regional plans are mandated to align with this objective. | 高 | SM005, SM006 |
| CM017 | China's manufacturing industry has an operational stock of approximately 2 million industrial robots—approximately 4.5 times Japan's stock—and accounted for 54% of annual global industrial robot installations in the most recent IFR World Robotics Report. | 高 | SM005, SM016 |
| CM018 | The share of local suppliers in China's domestic industrial robot installations increased from 30% in 2020 to 57% in 2024, with Chinese manufacturers supplying 59% of the global electronics industry robot base and 85% in the metal and machinery segment. | 高 | SM005, SM004 |
| CM019 | China's MIIT published its first national standard system for humanoid robots and embodied intelligence in February 2026, organized around six pillars: foundational standards, neuromorphic computing, limbs and components, system integration, application scenarios, and safety and ethics. | 中 | SM006, SM015 |
| CM020 | Goldman Sachs China survey respondents (January 2026) reported that humanoid robot platform iteration cycles shortened to approximately 6–8 months per generation in 2025–2026, attributed to 80–90% independent component design capability enabling compressed R&D and testing cycles. | 中 | SM007, SM004 |
| CM021 | Humanoid robot shipment volumes reported by Chinese manufacturers are not independently audited; Unitree claimed 5,500 units shipped in 2025 and Agibot claimed 5,168—counts that differ in methodology (pure bipedal vs. mixed-form including wheeled robots) and cannot be third-party verified. | 中 | SM017, SM020 |
| CM022 | A Morgan Stanley AlphaWise survey of Chinese C-suite executives in 2026 found that 62% are likely to adopt humanoid robots within the next three years; however, the sample skewed toward firms already utilizing robotics, making the figure potentially optimistic. | 中 | SM011, SM007 |
| CM023 | Only 23% of Chinese enterprise respondents in the Morgan Stanley survey expressed satisfaction with current humanoid robot products, citing dexterity, functionality, and price as the primary shortfalls; approximately 10% are currently evaluating or launching pilot projects. | 中 | SM011, SM012 |
| CM024 | The Morgan Stanley survey of Chinese enterprises established a clear price ceiling: 92% of respondents stated that humanoid robot unit cost must fall below 200,000 yuan (~$28,000 USD) for widespread commercial adoption to be feasible. | 中 | SM011, SM007 |
| CM025 | Automotive OEM manufacturing is the confirmed leading buyer segment for industrial humanoid robots: UBTECH's Walker S2 fleet was delivered to BYD, Geely, FAW-Volkswagen, Dongfeng Liuzhou Motor, and Foxconn in 2025. | 中 | SM014, SM016 |
| CM026 | Agibot's G2 humanoid robot was deployed in a live production line at Longcheer Technology's tablet manufacturing facility, reporting throughput of up to 310 units per hour, cycle times of 19–20 seconds, and a success rate exceeding 99.9%—the first confirmed 3C electronics dexterous manipulation deployment in China. | 中 | SM021, SM025 |
| CM027 | Logistics and warehouse operations represent the second key buyer segment; Goldman Sachs China survey data indicates customers require robots to achieve approximately 50% of human worker capacity to justify investment, resulting in approximately a 2-year payback period at prevailing Chinese logistics labor rates. | 中 | SM007, SM011 |
| CM028 | UBTECH's 2025 order book for its industrial Walker S2 humanoid robot exceeded 800 million yuan (~$113M USD), with buyers drawn from automotive and logistics sectors—the largest confirmed humanoid order book disclosed by any Chinese manufacturer as of June 2026. | 中 | SM014, SM020 |
| CM029 | Goldman Sachs' January 2026 China survey identified security patrols, hotel and museum guidance services, and simple logistics sorting as the near-term viable use cases because they require only mobility and basic interaction, not advanced dexterous manipulation. | 中 | SM007, SM011 |
| CM030 | Global robotics startups raised a record $14 billion in 2025, up from $8.2 billion in 2024, while China-based robotics companies raised $5.6 billion in venture investment through mid-May 2026 alone across 176 deals—sustaining the VC wave into 2026. | 高 | SM013, SM022 |
| CM031 | Agilink was spun out of Agibot in January 2026, raised four funding rounds, and reached a $1 billion valuation by May 2026, according to Crunchbase News; its business model involves dexterous hand hardware sales and licensing to the broader robotics market. | 高 | SM022, SM023 |
| CM032 | China's National Development and Reform Commission (NDRC) issued an explicit warning in 2026 against "highly repetitive products" in the humanoid robotics sector, noting that over 150 humanoid robot companies now operate in China. | 中 | SM012, SM015 |
| CM033 | The NDRC characterized "speed and bubbles" as perennial risks in frontier industries that must be "managed and balanced," signaling a policy shift from unbridled subsidy expansion toward selectivity and consolidation in the humanoid robotics sector. | 中 | SM012, SM017 |
| CM034 | The International Federation of Robotics assessed in May 2026 that "mass adoption as universal humanoid factory helpers or in private households will not happen within the near- and medium-term future," citing the superiority of traditional industrial robots for repetitive, high-speed, precision manufacturing. | 高 | SM005, SM010 |
| CM035 | Goldman Sachs' January 2026 China industry survey found the sector has pivoted from "universal imagination" to "dedicated landing" in narrow use cases; general-purpose dexterous manipulation at commercial scale was assessed as 1–3 years away from viability as of early 2026. | 中 | SM007, SM011 |
| CM036 | The sim-to-real gap remains a confirmed bottleneck for humanoid robot deployment: robot AI accuracy achieves 80–90% in simulation but drops below 50% in real-world environments, with dexterous manipulation tasks most exposed to this failure mode due to their high tactile- sensing fidelity requirements. | 高 | SM007, SM004 |
| CM037 | The American Security Robotics Act (ASRA), introduced by Senators Tom Cotton and Chuck Schumer with a companion House bill from Representative Stefanik in March 2026, proposes banning U.S. federal government procurement of Chinese-made humanoid robots; Agibot and Unitree are explicitly cited in the legislative record. | 高 | SM018, SM006 |
| CM038 | No independent analyst report with a standalone dexterous robotic hand market sizing as a discrete category was publicly accessible as of June 2026; all cited market figures are proxies for the broader humanoid hardware segment. | 中 | |
| CM039 | Goldman Sachs estimates humanoid robot solutions could become economically viable in factory settings between 2025 and 2028, and in consumer applications between 2030 and 2035, requiring approximately 15–20% annual production cost reductions each year. | 高 | SM002, SM007 |
| CM040 | Factory collaborative robots ("cobots") took approximately 7–10 years from first commercially available versions to batch-sale viability, facing similar skepticism about dexterity, safety, and integration as humanoid robots face today—providing an industry adoption-curve analog. | 高 | SM010, SM002 |
| CM041 | Unitree Robotics reported 2025 revenue of over 1.7 billion RMB (~$235M USD), a 335% year-on-year increase, with humanoid robots now accounting for over 51% of core income, providing the most concrete proxy for the commercial-scale humanoid hardware market in China. | 中 | SM020, SM013 |
| CM042 | China's MIIT Standards Technical Committee for Humanoid Robots comprises 65 members led by MIIT government officials with Agibot's co-founder and Unitree's founder as deputy chairs, creating the potential for published standards to embed specifications that advantage committee-affiliated supply chains including Agilink. | 中 | SM015, SM006 |
| CP001 | The dexterous robotic hand competitive landscape consists of five categories: Western research incumbents, academic open-source platforms, Chinese commercial standalone hands, platform-bundled OEM hands, and adjacent logistics end-effector systems. | 高 | SP001, SP005, SP003, SP006, SP009 |
| CP002 | RightHand Robotics RightPick 3 and Dexterity AI Mech are system-level logistics competitors that substitute for standalone dexterous hands in high-throughput fulfillment workflows. | 中 | SP009, SP011 |
| CP003 | LEAP Hand, published at RSS 2023 by CMU researchers, is the primary active open-source dexterous hand platform used as a research baseline, distributed under MIT license at approximately $2,000 assembly cost. | 高 | SP003, SP004 |
| CP004 | Fourier Intelligence GR-2 and Unitree R1 are humanoid OEM platforms that bundle proprietary dexterous end effectors, competing with Agilink's standalone offering for embodied-AI deployment budgets without selling the hand as a discrete SKU. | 中 | SP007, SP014 |
| CP005 | The status-quo alternative for manufacturing and logistics buyers remains human labor supplemented by fixed automation (conveyor systems, vacuum suction, two-jaw grippers), setting the economic bar a dexterous hand must clear in real deployments. | 中 | SP010, SP011 |
| CP006 | Shadow Robot Company (London, founded 1987) has built the Shadow Dexterous Hand with 24 DOF, holding the highest publicly documented DOF count among commercial dexterous hands and a Humanoid Index readiness score of 8/10. | 中 | SP001, SP002 |
| CP007 | Shadow Robot's list pricing for the Shadow Dexterous Hand system is not publicly disclosed on any accessible web page; buyers must request a quote directly. | 中 | SP001, SP002 |
| CP008 | Allegro Hand V4 (Wonik Robotics, South Korea) offers 16 DOF at a list price of $24,500, weighing approximately 1.1 to 1.2 kg, with no integrated tactile or force sensors. | 中 | SP005 |
| CP009 | LEAP Hand (CMU, RSS 2023) was verified in a peer-reviewed conference paper to assemble in four hours at a cost of $2,000 using off-the-shelf parts and is released under MIT license. | 高 | SP003, SP004 |
| CP010 | LEAP Hand's Python, C++, and ROS/ROS2 API supports query rates of up to 500 Hz, and the hand outperformed Allegro Hand in all benchmark manipulation experiments reported in its RSS 2023 paper. | 高 | SP003, SP004 |
| CP011 | LEAP Hand outperforms Allegro Hand at one-eighth of the cost because its novel kinematic structure maximizes dexterity regardless of finger pose, whereas Allegro's DC motor torque control degrades at extreme joint angles. | 中 | SP003 |
| CP012 | For academic buyers whose primary criterion is data-collection throughput at low cost, LEAP Hand represents a credible substitute for any commercial dexterous hand priced below $15,000, including OmniHand O10. | 中 | SP003, SP005, SP020 |
| CP013 | Inspire Robots (Beijing Yinshi Robot Technology Co., founded 2016) manufactures precision micro-servo actuators and dexterous hands for the Chinese domestic industrial market, with product lines including RH56BFX, RH56DFX, RH56E2, RH56F1, RH5DG2, and EG2 series grippers. | 中 | SP012 |
| CP014 | Inspire Robots' English-language product documentation and subpages were not accessible (returning 404 errors) as of the June 2026 fetch, limiting independent international due diligence on specifications and pricing. | 中 | SP012 |
| CP015 | RobotEra XHAND1 offers 12 active DOF, 25 kg grip force, 1.1 kg weight, integrated tactile and force sensing, gear-driven force-controlled joints, and a list price of $14,000 as documented by Humanoid.guide. | 中 | SP006, SP016 |
| CP016 | RobotEra publicly describes its XHAND1 full direct-drive architecture as 'the first of its kind in the industry,' enabling high-precision, adaptable, and durable manipulation validated through long-term real-world deployment. | 中 | SP016 |
| CP017 | RobotEra's XHAND1 and overall hardware system have been adopted by Boston Dynamics, NVIDIA, and Apple as confirmed by RoboticsTomorrow reporting on the May 2026 funding announcement. | 中 | SP016, SP017 |
| CP018 | RobotEra raised over $200 million in a new round led by SF Group (logistics giant), plus a prior RMB 1 billion ($146M) strategic round closed approximately one month earlier, bringing its disclosed 2026 total above $340 million. | 中 | SP016, SP018 |
| CP019 | RobotEra initiated thousand-unit deliveries in Q2 2026 and deployed robots across more than 10 logistics centers in collaboration with China Post and SF Express, with reported annualized growth of 300%. | 中 | SP016, SP018 |
| CP020 | RobotEra founder Chen Jianyu stated that nine of the world's ten most valuable technology companies are clients of RobotEra's hardware, though this claim has not been independently verified in a separate corroborating source. | 低 | SP017 |
| CP021 | Fourier Intelligence GR-2 offers 53 total DOF with 12 DOF per hand and six array-type tactile sensors per hand, priced at approximately $100,000 or more for the full system, with 500+ units shipped in 2025. | 中 | SP007, SP008 |
| CP022 | Fourier Intelligence is SoftBank-backed, has raised over $200M at an $800M+ valuation, employs 800+ people, and holds a Series D (2024) funding status, making it the best-resourced Chinese dexterous-hand competitor to Agilink by funding and headcount. | 中 | SP007, SP008 |
| CP023 | Unitree released the R1 dual-arm humanoid in April 2026 at a starting price of ¥26,900 ($3,930), with the R1 model priced at $5,900 (25 kg) and swappable end effectors that include two-, three-, and five-finger dexterous hand options. | 高 | SP014, SP015 |
| CP024 | Unitree's humanoid robot business became its largest revenue segment in 2025, driving a 335.4% annual revenue surge, and the company has filed for an IPO on Shanghai's STAR Market to raise up to ¥4.2 billion. | 中 | SP015 |
| CP025 | Unitree's R1 supports 15 to 31 configurable DOF in the arm assembly (5×2 or 7×2 arm configurations) and provides a 10 TOPS base compute module, expandable to 40–100 TOPS with optional NVIDIA Jetson Orin. | 中 | SP015 |
| CP026 | Dexterity AI's 'Foresight' world model has trained on over 100 million autonomous actions deployed in production environments and powers safety-rated physical AI for Fortune 50 enterprise customers. | 低 | SP011 |
| CP027 | Dexterity AI began operational validation of its Mech superhumanoid for truck-loading at Sagawa Express in Japan in October 2025, and achieved a 17× end-to-end inference speedup on NVIDIA TensorRT. | 中 | SP011 |
| CP028 | RightHand Robotics' RightPick 3 achieves up to 1,200 picks per hour with Level 4 autonomy and greater than 99% pick accuracy, handling items up to 2 kg, as documented in a named deployment at PALTAC Japan (500,000 sq ft facility) and Apologistics Netherlands. | 中 | SP009, SP010 |
| CP029 | RightHand Robotics' last disclosed funding event was its $23 million Series B round closed in December 2018, followed by $6 million in debt financing in July 2020; no subsequent equity funding rounds were publicly disclosed as of June 2026. | 中 | SP009 |
| CP030 | RightHand Robotics RightPick 3 is not an anthropomorphic dexterous hand but a task-specific picking system that eliminates the need for a multi-DOF hand entirely within the piece-picking use case. | 中 | SP009, SP010 |
| CP031 | RightHand Robotics RightPick 3 uses a UR5e or UR10e cobot arm as the standard base, with a dual-GPU industrial compute platform (RightPick processor) and includes fleet management and real-time performance dashboards. | 中 | SP009 |
| CP032 | Dexterity AI partnered with Beckhoff USA (EtherCAT inventor) in November 2025 and with ASRock Rack for physical AI supercomputers, deepening its enterprise-automation integration beyond what standalone hand vendors offer. | 中 | SP011 |
| CP033 | OmniHand O10 (Agilink) lists at $3,775 on Humanoid.guide, which is 85% below Allegro Hand V4's $24,500 list price, while matching Allegro's 16 DOF count and adding tactile sensing Allegro does not have. | 中 | SP020, SP005 |
| CP034 | OmniHand Pro (O12) at $12,700 offers 19 DOF, 35 kg grip force, 0.75 kg weight, and multi-axis force sensing; OmniHand O10 at $3,775 offers 16 DOF, 2 kg grip force, 0.5 kg weight, and 400+ tactile sensor points. | 中 | SP020, SP021 |
| CP035 | OmniHand 3 Ultra-T features a 22+3 DOF tendon-driven system, 500g weight, 10:1 load-to-weight ratio, full-hand 3D tactile sensing, an integrated palm camera, sub-0.3-second response, and a wrist range of 55° pitch and 40° yaw. | 中 | SP019 |
| CP036 | Agilink's Genie Studio developer platform supports one-click robot deployment, achieving 2–3× performance gains over traditional single-GPU inference, and supports foundation models including RDT, Pi0, OpenVLA, GR00T, and AgiBot Go-1. | 中 | SP022 |
| CP037 | AimRT, the open-source middleware underpinning Agilink and Agibot deployments, reached version 1.7.0 in April 2026 with ROS2 Jazzy support, zenoh plugin, gRPC, MQTT, and MCAP recording capabilities. | 中 | SP023 |
| CP038 | Agibot reported its 10,000th unit milestone in March 2026 and declared 2026 'Deployment Year One,' providing Agilink with a parent-ecosystem reference base that no independent dexterous hand company currently matches in China. | 中 | SP024, SP019 |
| CP039 | RobotEra's direct-drive XHAND1 being adopted by Boston Dynamics, NVIDIA, and Apple as hardware creates a credibility signal that Agilink's tendon-driven OmniHand 3 Ultra-T cannot yet match with published third-party validation. | 中 | SP016, SP017 |
| CP040 | U.S. lawmakers introduced a bill in March 2026 to ban government procurement of Chinese-manufactured robots, according to reporting cited in earlier chapters; this legislative headwind applies to Agilink, RobotEra, Unitree, and all PRC-origin robotic hand vendors equally. | 中 | SP024 |
| CP041 | Unitree R1's complete system at $3,930 creates a perception problem for Agilink's standalone hand pricing: a buyer seeking dexterous manipulation capability can acquire a full dual-arm platform for approximately the same price as OmniHand O10 alone. | 中 | SP014, SP015, SP020 |
| CP042 | Platform bundling by Unitree, Fourier, and RobotEra — where the dexterous hand is sold only as part of the full humanoid system — structurally reduces the total addressable market for standalone dexterous hand companies like Agilink over time. | 中 | SP007, SP014, SP016 |
| CP043 | OmniHand O10 and Unitree R1's dexterous end-effector option cluster in the high commercial maturity, mid-DOF quadrant of the competitive landscape, while Shadow Robot leads on DOF with lower commercial deployment scale. | 中 | SP001, SP014, SP020 |
| CP044 | OmniHand 3 Ultra-T with 22+3 DOF and RobotEra's direct-drive XHAND1 at 12 DOF represent competing technical philosophies: higher-DOF tendon-driven dexterity versus lower-DOF direct-drive durability and precision; neither had published a head-to-head benchmark as of June 2026. | 中 | SP019, SP016 |
| CP045 | No competitor in the Chinese commercial dexterous hand segment publicly combines more than 16 DOF, tactile sensing, a palm camera, and a commercial AI deployment platform in a standalone sub-1 kg product below $15,000 — Agilink's OmniHand 3 Ultra-T is the only product that meets all five criteria simultaneously, though its commercial pricing has not been disclosed. | 中 | SP006, SP019, SP020 |
| CP046 | OmniHand O10's $3,775 price represents an 85% discount to Allegro Hand V4's $24,500 price for equivalent DOF count, making it the most price-competitive tactile-enabled commercial dexterous hand at its specification tier. | 中 | SP005, SP020 |
| CP047 | Agibot (Agilink's parent) shipped more than 5,100 humanoid units in 2025, ranked #1 globally by Omdia, providing Agilink with unmatched production scale reference among Chinese dexterous hand companies. | 中 | SP024 |
| CP048 | RobotEra's combined 2026 disclosed funding exceeded $340 million (RMB 1 billion strategic round plus $200 million new round), making it the best-capitalized pure-play dexterous hand company active in China as of June 2026. | 中 | SP016, SP018 |
| CI001 | Agilink's revenue model is hardware-first, anchored to per-unit sales of dexterous hands and grippers with no publicly disclosed recurring software or services revenue as of June 2026. | 中 | SI011, SI013, SI016 |
| CI002 | Agibot (Zhiyuan) retained approximately 80% of Agilink's equity at the time of its January 2026 incorporation. | 中 | SI020, SI027 |
| CI003 | Agilink completed four funding rounds between January and May 2026, all within approximately five months of its incorporation. | 高 | SI001, SI011, SI021 |
| CI004 | Agilink's first two funding rounds (January 2026) were led by Hillhouse Venture and BlueRun Ventures. | 中 | SI020, SI023 |
| CI005 | Agilink's Round 3 (announced February 10–11, 2026) raised several hundred million yuan and was led by a major internet company. | 中 | SI003, SI023 |
| CI006 | Round 3 investors include Baidu Ventures, Yunfeng Capital (Yunfeng Fund LP), C Capital, Synstellation Capital, Joyson Electronics (Ningbo Joyson Electronic Corp.), Hangzhou Longqi Scientific Investment (Longcheer Technology), SAIC Motor Financial Holdings, BRV Partners, Hillhouse Capital, and BlueRun Ventures. | 中 | SI003, SI004 |
| CI007 | C Capital confirmed its participation in Agilink's multi-hundred-million RMB financing round, noting it is an early investor in Agibot since 2023 and views Agilink as a key embodied intelligence component play. | 中 | SI004 |
| CI008 | Agilink completed a fourth funding round in May 2026, raising additional several hundred million yuan. | 中 | SI011, SI021, SI022 |
| CI009 | Confirmed investors in Agilink's fourth round (May 2026) include Baidu Ventures, Yunfeng Capital, SAIC Financial Holdings, and Hillhouse Capital. | 中 | SI011 |
| CI010 | Agilink's post-money valuation exceeded $1 billion USD (unicorn threshold) following its fourth funding round in May 2026. | 高 | SI001, SI010, SI011, SI021 |
| CI011 | Agilink achieved unicorn status in fewer than 150 days after its January 2026 incorporation, described by ITJuzi analyst Wu Meimei as unprecedented in the humanoid component sector. | 高 | SI001, SI011 |
| CI012 | Agilink's total lifetime capital raised across all four rounds has not been publicly disclosed as of June 2026. | 中 | SI011, SI003 |
| CI013 | The OmniHand 2025 (O10) carries a publicly confirmed list price of 9,800 RMB (approximately $1,350 USD), making it the first high-DoF dexterous hand priced under 10,000 RMB. | 中 | SI014, SI015 |
| CI014 | Before Agilink's OmniHand 2025, commercial dexterous hands typically started at $10,000 USD (approximately 73,000 RMB) and reached up to $200,000 for research-grade platforms. | 中 | SI014 |
| CI015 | The OmniHand Pro 2025 is available only via enterprise quotation with no publicly disclosed list price. | 中 | SI014, SI017 |
| CI016 | Third-party reseller Botonly lists the OmniHand O10 at $4,420 USD as of June 1, 2026, a roughly 3.3x premium over Agibot's mainland China list price of 9,800 RMB (~$1,350 USD). | 中 | SI015 |
| CI017 | According to Xiong Kun (Agilink CEO), the dexterous hand market is divided into three tiers: entry (6–10 DoF, several thousand to 30,000 RMB), mid-range (11–17 DoF, 50,000–100,000+ RMB), and high-end (20+ DoF, 200,000+ RMB). | 中 | SI005 |
| CI018 | The OmniHand 3 Lite is described by Humanoids Daily as 'aggressively priced' but no specific price has been disclosed at launch. | 低 | SI011 |
| CI019 | Agilink delivered more than 8,000 dexterous hands cumulatively as of approximately May 19, 2026. | 中 | SI011, SI013, SI018 |
| CI020 | Agilink delivered more than 10,000 grippers cumulatively as of approximately May 19, 2026. | 中 | SI011, SI013 |
| CI021 | Approximately 1,000 of Agilink's delivered units were actively deployed in 8-hour continuous shifts in retail warehouses, pharmacies, and factories as of May 2026. | 低 | SI011 |
| CI022 | At the OmniHand 2025 floor ASP of 9,800 RMB, 8,000 delivered hands imply cumulative hardware revenue of approximately 78.4 million RMB (~$10.8 million USD). | 低 | SI013, SI014, SI019 |
| CI023 | Incorporating a product mix of O10, OmniHand Pro, and gripper units, estimated cumulative Agilink revenue ranges from approximately 50 million to 200 million RMB ($7–27 million USD) as of May 2026. | 低 | SI014, SI015, SI011 |
| CI024 | No revenue, ARR, gross margin, burn rate, or cash position has been publicly disclosed by Agilink as of June 2026. | 中 | SI011, SI013, SI021 |
| CI025 | Dexterous hands represent approximately 20% of a humanoid robot's total bill of materials, making them a structurally important sub-component in the humanoid supply chain. | 低 | SI011 |
| CI026 | For Chinese mid-range dexterous hands, estimated BOM composition is: motors 30–40%, sensors and electronics 20–30%, structural materials 20–25%, cables and gears 5–10%, assembly and labor 5–15%. | 低 | SI014, SI005 |
| CI027 | Agilink self-develops key motors and materials, which Xiong Kun cited as the basis for cost control and 'core pricing power' versus assembly-dependent competitors. | 低 | SI005 |
| CI028 | Agilink benefits from access to Agibot's supply chain infrastructure, reducing component costs and accelerating manufacturing scale versus a standalone startup. | 低 | SI005, SI012 |
| CI029 | China's embodied-AI sector recorded more than 50 financing events in Q1 2026, channeling approximately 20 billion yuan ($2.94 billion) into the space, a nearly 60% year-on-year increase. | 中 | SI011 |
| CI030 | China's NDRC warned in 2025 that 150+ humanoid robot companies risk 'speed and bubbles' through repetitive products and inadequate differentiation. | 高 | SI002, SI006, SI019 |
| CI031 | More than half of China's 150+ humanoid robot companies are startups or cross-industry entrants, creating market conditions for homogeneous products that compress R&D investment capacity. | 高 | SI002, SI006 |
| CI032 | Agilink's $1B+ valuation versus estimated cumulative revenue of $7–27 million implies a speculative valuation multiple in excess of 40x, indicating investors are pricing in scale optionality rather than current earnings. | 低 | SI001, SI011 |
| CI033 | Agibot (Agilink's parent) signed a commercial agreement with Hyperscale Data's Omnipresent Robotics in May 2026 to supply up to 143 robots for US deployment, evidencing Agibot ecosystem commercialization momentum that is contextually relevant to Agilink's supply chain positioning. | 中 | SI007, SI008 |
| CI034 | Agibot (Agilink's parent) is pursuing a STAR Market IPO, which, if completed, would create a public financial disclosure trail for the parent that could illuminate Agilink's supply-chain and related-party economics. | 中 | SI024, SI025 |
| CI035 | No independent board, auditor, liquidation preference terms, ESOP structure, or governance documentation has been publicly disclosed for Agilink as of June 2026. | 中 | SI011, SI021 |
| CI036 | Agibot's approximately 80% equity stake at formation, combined with its likely role as primary customer for Agilink hands, creates a material related-party and customer-concentration risk that requires investor-level disclosure to quantify. | 中 | SI002, SI009, SI020 |
| CI037 | Estimated aggregate lifetime capital raised by Agilink is above 500 million RMB (~$68 million USD), based on at least three of four rounds described as 'several hundred million yuan' or larger. | 低 | SI003, SI011, SI023 |
| CI038 | Agilink's stated planned use of funds from the fourth round includes: building a reliable general-purpose dexterous actuation platform, expanding mass production and delivery capacity, and developing full-stack AI model and data capabilities. | 中 | SI011 |
| CI039 | Agilink's monthly burn rate has not been disclosed; hardware manufacturing scale-up at this stage typically implies 15–40 million RMB per month for engineering, capex, and component procurement combined. | 低 | SI014, SI005 |
| CI040 | No debt facility, credit line, project finance obligation, or factoring arrangement has been publicly disclosed for Agilink as of June 2026. | 中 | SI011, SI021 |
| CI041 | Joyson Electronics' 2025 annual report, filed via the CSRC-mandated cninfo.com.cn platform, discloses strategic expansion into the 'emerging intelligent body supply chain,' consistent with its Round 3 investment in Agilink. | 中 | SI009 |
| CI042 | The OmniHand 2025 list price of 9,800 RMB represents a 7x–55x reduction relative to prior commercial dexterous hands ($10,000–$200,000 starting price), enabling Agilink to address a dramatically larger addressable customer base. | 中 | SI014 |
| CI043 | Agilink plans to present at ICRA 2026 in Vienna with its full dexterous manipulation portfolio, including the world premiere of the OmniHand 3 Ultra-M direct-drive model, indicating an active international commercial development effort. | 中 | SI011 |
| CI044 | Agilink's sales GTM (as inferred from product positioning) combines: direct web/store sales for the OmniHand 2025, enterprise-channel sales for the Pro and OmniHand 3 series, and internal supply to Agibot for humanoid robot integration — but no CAC, sales cycle, or channel split data has been publicly disclosed. | 低 | SI014, SI005, SI016 |
| CE001 | OmniHand 2025 has 16 total degrees of freedom (10 active, 6 passive) in a 180×85×38.5 mm five-finger anthropomorphic form factor. | 高 | SE001, SE005, SE007 |
| CE002 | OmniHand 2025 weighs ≤500 g in base configuration and ≤550 g with the optional tactile module installed. | 高 | SE001, SE007 |
| CE003 | OmniHand 2025 uses a linkage mechanism transmission driven by motor + gear at each active joint, enabling a 0.5 mm positional repeatability. | 高 | SE001, SE005 |
| CE004 | OmniHand 2025 maximum fingertip force is 5 N, with a minimum grasp diameter of 5 mm. | 高 | SE001, SE007 |
| CE005 | OmniHand 2025 minimum close time is 0.7 s (typical), enabling fast grasp and release cycles. | 中 | SE001 |
| CE006 | OmniHand 2025 supports both CANFD and RS485 communication interfaces at a 1 kHz communication rate; the Pro 2025 uses CANFD exclusively. | 高 | SE001, SE002, SE005 |
| CE007 | The optional tactile module for OmniHand 2025 provides 400+ force-sensitive contact points at 0.1 N array resolution covering a 0–20 N sensing range; the Pro has 150+ points at 0.01 N over 0–50 N. | 高 | SE001, SE002, SE005 |
| CE008 | OmniHand 2025 load capacity is 1 kg (typical) and 2 kg (maximum) palm up with a 5 cm rigid object. | 中 | SE001 |
| CE009 | OmniHand Pro 2025 has 19 total degrees of freedom (12 active, 7 passive) in a 207×98×56 mm form factor. | 高 | SE002, SE008, SE005 |
| CE010 | OmniHand Pro 2025 weighs ≤750 g, 250 g more than the base OmniHand 2025. | 高 | SE002, SE008 |
| CE011 | OmniHand Pro 2025 uses a linkage transmission driven by motor + screw rod, enabling higher load capacity at the cost of added weight versus the OmniHand 2025's motor + gear design. | 高 | SE002, SE005 |
| CE012 | OmniHand Pro 2025 maximum fingertip force is 20 N — four times the OmniHand 2025's 5 N maximum. | 高 | SE002, SE008 |
| CE013 | OmniHand Pro 2025 achieves a minimum grasp diameter of 2 mm and a positional repeatability of 0.02 mm (typical), enabling precision assembly of small electronic components. | 高 | SE002, SE005 |
| CE014 | OmniHand Pro 2025 load capacity is 35 kg (palm up, 5 cm rigid object) and 5 kg (palm down), far exceeding the OmniHand 2025's 2 kg maximum. | 高 | SE002, SE008 |
| CE015 | OmniHand Pro 2025 multi-modal sensing includes 3-axis fingertip force (detecting normal pressure and tangential shear), 1-axis palm force, and 150+ tactile points at 0.01 N resolution over 0–50 N range. | 高 | SE002, SE005 |
| CE016 | OmniHand Pro 2025 maximum non-destructive force is 1000 N, indicating structural resilience to inadvertent heavy contact without sensor damage. | 中 | SE002 |
| CE017 | Agibot company claims OmniHand 3 Ultra-T has 22+3 DoF (tendon-driven) at 500 g weight with a 55° pitch and 40° yaw wrist range, announced April 17 2026 at APC 2026. | 中 | SE003, SE006 |
| CE018 | Agibot company claims a 10:1 load-to-weight ratio for OmniHand 3 Ultra-T (500 g hand, ~5 kg payload); no independent third-party verification of this ratio has been published. | 低 | SE003 |
| CE019 | Agibot company claims OmniHand 3 Ultra-T integrates full-hand 3D tactile sensing and a palm camera; no sensor count, resolution, or sensing range has been published for this generation. | 低 | SE003, SE006 |
| CE020 | Agibot claims OmniHand 3 Ultra-T achieves sub-0.3-second response time for full-hand posture change; no test protocol, measurement methodology, or repeatability data has been disclosed. | 低 | SE003 |
| CE021 | Agibot company states OmniHand 3 Ultra-T wrist range is 55° pitch and 40° yaw, providing a wider articulation range than the base OmniHand 2025 which has no disclosed wrist specs. | 中 | SE003 |
| CE022 | OmniHand 3 Ultra-T was announced April 17 2026 with no disclosed commercial availability date, pricing, or delivery timeline as of June 16 2026. | 高 | SE003, SE006 |
| CE023 | OmniHand 2025 SDK supports Python 3.10+ and C++ APIs on Ubuntu 22.04 (x86_64) only, requiring CMake 3.24+ and gcc 11.4+ for source compilation. | 中 | SE004 |
| CE024 | OmniHand SDK CANFD communication requires a ZLG USBCANFD series adapter (recommended: USBCANFD-100U-mini, USBCANFD-100U, or USBCANFD-200U), creating a single-vendor hardware dependency for communication. | 中 | SE004, SE005 |
| CE025 | OmniHand SDK is open-sourced on GitHub under the Mulan Public Source License v2, a Chinese-government-backed permissive license compatible with Apache 2.0 for most commercial uses. | 中 | SE004, SE005 |
| CE026 | OmniHand 2025 SDK supports five control modes: P-A (position/angle), Torque, Hybrid PtA (position + torque), Velocity, and Tactile mode. | 高 | SE001, SE004 |
| CE027 | OmniHand 2025 launched in August 2025 at ¥9,800 (approximately $1,350 USD), with a temporary discount price of ¥5,000, positioning it as the first commercially available high-DoF hand below the ¥10,000 price point. | 中 | SE009, SE005, SE007 |
| CE028 | OmniHand Pro 2025 is listed at approximately $12,700 USD per humanoid.guide's distributor data. | 中 | SE008 |
| CE029 | Genie Studio supports RDT, Pi0, OpenVLA, GR00T, and AgiBot's proprietary Go-1 foundation models for manipulation policy training, with a claimed 48 GPU-days for new-scenario fine-tuning. | 中 | SE010 |
| CE030 | AgiBot World dataset (hosted on HuggingFace) contains 1M+ trajectories spanning 217 tasks, 87 skills, 3,000+ objects, and 100+ real-world scenarios across five domains, collected using dexterous hands including OmniHand variants. | 中 | SE019 |
| CE031 | LEAP Hand (Carnegie Mellon University) offers 16 DoF at $2,000 USD with 0.94 kg weight and servo motor drive, with approximately 2 kg payload, positioned as the open-source research reference for affordable dexterous manipulation. | 高 | SE014, SE015 |
| CE032 | Allegro Hand V4 (Wonik Robotics) offers 16 DoF with established ROS ecosystem support; Allegro V5 Sense extends to tactile sensing, but the line is priced at $15,000+ and historically oriented toward research rather than commercial humanoid deployment. | 中 | SE016, SE017, SE018 |
| CE033 | XHAND1 (Robotera) offers 12 DoF, $14,000 USD, 1.1 kg weight, 25 kg load capacity, and gear-driven force-controlled joint modules, positioning it as a direct industrial-tier competitor to OmniHand Pro 2025. | 中 | SE013 |
| CE034 | The human hand has approximately 27 degrees of freedom and over 17,000 mechanoreceptors distributed across the palm and fingers, providing a biological capability benchmark that commercial dexterous hands remain far below in tactile sensing density. | 中 | SE020 |
| CE035 | Both OmniHand 2025 and Pro 2025 SDKs include URDF models for simulation environments and ROS 2 driver packages for middleware integration; OmniHand Pro SDK also supports field- swappable modular fingertips for ESD-sensitive or standard manufacturing contexts. | 中 | SE005, SE004 |
| CE036 | OmniHand 2025 size of 180×85×38.5 mm and ≤500 g weight allow direct mounting to standard humanoid robot wrist interfaces and collaborative robot arms without structural reinforcement. | 中 | SE005, SE001 |
| CE037 | OmniHand Pro 2025 operating temperature range is -20°C to 50°C, compared to -10°C to 45°C for OmniHand 2025, suggesting the Pro is more suited to industrial outdoor or cold environments. | 高 | SE002, SE001 |
| CE038 | OmniHand Pro 2025 supports field-swappable modular fingertip pads allowing operators to change friction, ESD behavior, or texture without redesigning the end-effector tooling. | 中 | SE005 |
| CE039 | OmniHand Pro 2025 was demonstrated in live production at the Agibot G2 industrial humanoid launch (October 2025) for logistics parcel sorting and precision assembly tasks including RAM module insertion and seatbelt lock cylinder assembly. | 中 | SE005, SE006 |
| CE040 | OmniHand 2025 SDK version 0.8.0 supports Ubuntu 22.04 x86_64 only; no ARM (Jetson, CM4) or Windows deployment path is documented, limiting edge and embedded use cases. | 中 | SE004 |
| CE041 | As of June 16 2026, Agilink does not operate an independent product domain (agilink.cn returns a DNS failure); all OmniHand product pages remain on agibot.com, blurring the brand and IP boundary between the parent and spinout. | 中 | SE022, SE003 |
| CE042 | No MTBF, cycle-durability, warranty terms, or field failure rate data have been publicly disclosed for any OmniHand variant, making component lifetime and total cost of ownership analysis speculative. | 中 | |
| CE043 | OmniHand SDK version 0.8.0 pre-1.0 label implies it may undergo breaking API changes; no SDK lifecycle or long-term support commitment has been published by Agilink or Agibot. | 中 | SE004 |
| CU001 | Longcheer Technology (Nanchang, Jiangxi) is the sole publicly named, production-scale external customer of the Agibot G2 robot system equipped with OmniHand, with independently corroborated performance metrics. | 高 | SU001, SU003, SU014 |
| CU002 | Agibot G2 robots achieved up to 310 units per hour (UPH) throughput at Longcheer's MMIT (Multimedia Integrated Testing) tablet assembly stations. | 高 | SU001, SU002, SU003 |
| CU003 | Agibot G2 robots completed 2,283 tasks without a single error during an 8-hour continuous operation run at Longcheer's tablet factory in Nanchang. | 中 | SU002, SU003 |
| CU004 | The motion success rate in continuous operation at Longcheer exceeded 99.9%, with downtime loss below 4% over 140+ cumulative hours. | 中 | SU001, SU003 |
| CU005 | Production line integration at Longcheer was completed within 36 hours using the Genie Sim 3.0 digital twin platform; total project ramp from initiation to deployment took four months. | 中 | SU001, SU002, SU003 |
| CU006 | Agibot's stated target is to scale to 100 G2 robots at Longcheer by Q3 2026, with planned expansion into automotive and semiconductor sectors. | 低 | SU001, SU002 |
| CU007 | The Longcheer deployment supports 24/7 autonomous operation with minimal human intervention and handles mixed-model production without custom tooling; model changeover retraining takes ≤4 hours. | 中 | SU001, SU003 |
| CU008 | Li Long, General Manager of Longcheer's Robotics Division, provided an independent on-record quote in the official press release corroborating the four-month integration timeline and production-ready performance of the Agibot G2 deployment. | 高 | SU001, SU003 |
| CU009 | Longcheer Technology is a global smart-device ODM manufacturer with seven R&D centers (Shanghai, Shenzhen, Huizhou) and four major manufacturing hubs (Huizhou, Nanchang, Vietnam, India), covering smartphones, AI PCs, automotive electronics, tablets, and wearables. | 高 | SU001, SU013 |
| CU010 | HT-Tech deployment processes up to one million chips per day in semiconductor chip loading/unloading operations, maintaining disk drop rates below 0.001%. | 低 | SU004 |
| CU011 | Intel is named in Agibot's Seven Solutions APC 2026 material as an enterprise deployment partner, processing up to 100,000 chips per day; this claim originates solely from Agibot's own marketing with no Intel-sourced confirmation. | 低 | SU004 |
| CU012 | SAIC-GM is named in the Seven Solutions presentation as an automotive deployment using Agibot robots for battery cell handling, achieving sub-second perception latency and near-continuous throughput. | 低 | SU004 |
| CU013 | FULIN Precision deploys Agibot robots for industrial material handling, transporting hundreds of standardized containers per shift with sustained multi-hour operation. | 低 | SU004 |
| CU014 | Damon logistics sorting deployment achieves up to 70% of human-level throughput, with second-pass success rates reaching 99%. | 低 | SU004 |
| CU015 | Xiaonan Intelligence irregular soft-package sorting deployment reduced labor costs by 50% and reaches processing speeds up to 435 items per hour. | 低 | SU004 |
| CU016 | Anta Sports store deployments drove a 20% increase in foot traffic and double-digit conversion gains in retail assistant use cases. | 低 | SU004 |
| CU017 | Haidilao restaurants deployed Agibot interactive queueing and guidance systems, improving customer flow and engagement. | 低 | SU004 |
| CU018 | Guangzhou Metro deployments reduced operational costs by 13% and resolved up to 80% of passenger inquiries through self-service interaction. | 低 | SU004 |
| CU019 | Shanghai Hongqiao International Airport cleaning deployment reduced overall costs by 18% while improving cleaning frequency and quality in large-area environments. | 低 | SU004 |
| CU020 | State Grid security patrol deployments improved inspection efficiency 'multiple times' over manual methods and reduced maintenance costs, operating continuously across complex terrains. | 低 | SU004 |
| CU021 | Genting Malaysia Bhd signed an MOU with Agibot on April 17, 2026, to integrate humanoid agents into Resorts World Genting across theme parks, hospitality, and entertainment, including a planned Robotics Gala Performance in early 2027. | 中 | SU008, SU005 |
| CU022 | Hyperscale Data (via Omnipresent Robotics) partnered with Agibot for a 100,000 sq ft Michigan facility dedicated to robot hardening for US industrial standards and teleoperation data collection, targeting 500 tech jobs. | 低 | SU008 |
| CU023 | Singtel signed a Robots-as-a-Service (RaaS) deal with Agibot to offer humanoid robots via 5G in Singapore, alongside an MOU with NCS (Singtel's tech arm) for smart building and public sector roles. | 低 | SU008 |
| CU024 | SIR Spa (Italian system integrator) is handling on-the-ground integration of Agibot A2 and X2 series robots into Italy's manufacturing and automotive sectors, supported by a manufacturing alliance with Minth Group. | 低 | SU008 |
| CU025 | Whale Cloud (Alibaba subsidiary) provides infrastructure for Agibot international deployment across 150+ markets, supporting telecom inspection and government service automation use cases. | 低 | SU008 |
| CU026 | Agibot's commercial fleet spans 17 countries as of APC 2026 (April 2026), reflecting deployments and partner presence in both Asia-Pacific and Europe. | 中 | SU007, SU018 |
| CU027 | Agibot shipped 5,100 humanoid units in 2025, ranked #1 globally by Omdia research, though Unitree disputes this ranking with a competing 5,500-unit claim. | 中 | SU009, SU021 |
| CU028 | Agibot reached its 10,000th cumulative humanoid unit in March 2026, scaling from 5,000 to 10,000 units in three months — a 4× acceleration versus the prior production phase. | 中 | SU007, SU018 |
| CU029 | Agibot's eight declared commercial application categories are: reception/hospitality, entertainment, industrial manufacturing, logistics sorting, security inspection and patrol, data collection and training, scientific research and education, and industrial/commercial cleaning. | 中 | SU006, SU018 |
| CU030 | Pandaily's headline reported Agilink shipped over 8,000 dexterous hands by Q1 2026, but the article body was inaccessible (js-only rendering); this figure cannot be independently verified from the accessible source text. | 低 | |
| CU031 | OmniHand is explicitly named as Agibot's dexterous manipulation system in the CES 2026 portfolio press release alongside the G2 Series industrial robots; the 'Powered by AGIBOT' customization program allows partners to configure hardware, implying a third-party OEM channel. | 中 | SU006 |
| CU032 | RobotEra's founder Chen Jianyu claimed nine of the world's ten most valuable tech companies among Robot Era's clients as of mid-2025, specifically naming Haier Smart Home, Lenovo, and BZS Technology Development; the broader 'top-10' claim is unverifiable. | 低 | SU010 |
| CU033 | RobotEra achieved deployments in more than ten logistics centers through collaboration with China Post and SF Group in Q2 2026, with thousand-unit deliveries initiated and growth exceeding 300%. | 中 | SU011 |
| CU034 | RightHand Robotics deployed its RightPick 3 system in production at PALTAC (500,000 sqft cosmetics/pharma facility, Japan) and Apologistics (pharma AS/RS, Netherlands), achieving 1,200 picks/hour with 99%+ accuracy — a higher-quality named customer disclosure than Agibot/Agilink's current public record. | 中 | SU012 |
| CU035 | Morgan Stanley's AlphaWise survey found only 23% of Chinese enterprise decision-makers satisfied with current humanoid robot products, with dexterity, functionality, and unit cost (threshold: ¥200,000 / ~$28,000) cited as the primary adoption barriers. | 高 | SU015, SU025 |
| CU036 | Only approximately 10% of surveyed Chinese enterprise executives are currently evaluating or launching humanoid robot pilot projects, confirming that actual deployment remains early-stage despite high stated adoption intent (62%). | 中 | SU015 |
| CU037 | The American Security Robotics Act (introduced by Senators Cotton and Schumer, March 2026) would prohibit US federal government procurement and operation of humanoid robots made by Chinese firms, citing data privacy and national security risks; Agibot and Unitree are named in the Reuters reporting. | 中 | SU016 |
| CU038 | Unitree Robotics disputes Agibot's IDC-ranked #1 global humanoid shipment title, claiming 5,500 units shipped by end-2025 versus Agibot's 5,100; neither claim has been independently audited by a third-party verifier. | 中 | SU021, SU009 |
| CU039 | No publicly documented evidence exists of any third-party OEM, research institution, or system integrator purchasing OmniHand as a standalone product from Agilink separately from the Agibot G2 robot bundle; all identified customer relationships are mediated through Agibot's full-robot platform. | 中 | SU017, SU020 |
| CU040 | Transfer pricing and revenue attribution between Agilink (the dexterous-hand spinout) and parent Agibot for OmniHand units embedded in G2 robots are entirely undisclosed; Agilink's standalone revenues, margins, and financials are unknown from public sources. | 中 | SU022, SU009 |
| CR001 | Two US senators (Cotton and Schumer) introduced the American Security Robotics Act (ASRA) on March 26, 2026, a bipartisan bill to ban US government agencies from buying or operating humanoid robots made by Chinese firms. | 中 | SR009 |
| CR002 | The ASRA bill explicitly named Agibot and Unitree as Chinese companies whose humanoid robots would be banned from US federal procurement. | 中 | SR009 |
| CR003 | BIS issued a rule in January 2025 requiring a licence to export advanced computing items to entities headquartered in Country Group D:5 (which includes China) or Macau, or with an ultimate parent headquartered in those countries. | 中 | SR001 |
| CR004 | BIS extended the Authorized IC Designer timeline until December 31, 2026, following publication of the advanced computing IC rule on January 16, 2025. | 中 | SR001 |
| CR005 | The EU AI Act classifies AI safety components in critical infrastructure — including robots whose failure could put life and health of citizens at risk — as high-risk AI systems. | 高 | SR003, SR002 |
| CR006 | High-risk AI systems under the EU AI Act must undergo conformity assessment, maintain high-quality datasets, log activity, provide detailed documentation, and secure approval from a notified body before being placed on the EU market. | 高 | SR003, SR002 |
| CR007 | OSHA has confirmed there are currently no specific OSHA standards for the robotics industry. | 高 | SR004, SR016 |
| CR008 | ISO 10218-1:2011, the primary industrial robot safety standard, was formally withdrawn and reached Withdrawal stage (95.99); it also explicitly did not apply to non-industrial robots. | 高 | SR006, SR016 |
| CR009 | ISO/TS 15066:2016 covers collaborative robots but is a technical specification rather than a mandatory safety standard and does not specify force limits for multi-DOF dexterous hands. | 中 | SR008 |
| CR010 | China's MIIT published its first national standard system for humanoid robots and embodied intelligence in late February 2026, covering six pillars including safety and ethics. | 中 | SR024, SR011 |
| CR011 | Agibot co-founder Peng Zhihui sits on the 65-member MIIT Standards Technical Committee for humanoid robots as deputy chair alongside Unitree's Wang Xingxing. | 中 | SR011 |
| CR012 | OFAC maintains sanctions lists covering individuals and entities in targeted countries; any secondary sanctions expansion to Chinese robotics sectors would restrict capital flows to Agilink. | 低 | SR005 |
| CR013 | Hill Dickinson's 2026 legal analysis identifies GDPR tension for humanoid robots that collect facial and biometric data, including questions about cloud-side data storage responsibility. | 中 | SR002 |
| CR014 | China and the Middle East are accelerating robot deployment facing fewer regulatory constraints, while Western jurisdictions may adopt more slowly due to stricter privacy laws and labour protections. | 中 | SR002 |
| CR015 | Hill Dickinson identifies the question of who is responsible if a robot causes harm — manufacturer, operator, or software provider — as legally unsettled as of 2026. | 中 | SR002 |
| CR016 | Robot AI accuracy drops from 80-90% in simulated environments to below 50% in real-world deployment scenarios, per IDC's 2026 humanoid commercialisation report. | 中 | SR015, SR022 |
| CR017 | IDC identifies the sim-to-real gap as a supply chain bottleneck for humanoid robotics commercialisation in 2026. | 中 | SR015 |
| CR018 | Goldman Sachs' January 2026 China survey identifies the time required for large-scale, high-quality real data collection as a primary factor limiting deployment speed. | 中 | SR022 |
| CR019 | Only 23% of Chinese enterprise C-suite respondents were satisfied with current humanoid robot products in Morgan Stanley's 2026 AlphaWise survey. | 中 | SR012 |
| CR020 | Dexterity and functionality were identified as the primary complaints with current-generation humanoid products in the Morgan Stanley survey. | 中 | SR012 |
| CR021 | Agibot G2 robots embedding OmniHand achieved 99.9%+ success rate at 310 UPH at Longcheer Technology's Nanchang plant — the only confirmed quantified production deployment. | 中 | SR019 |
| CR022 | The Longcheer deployment is a single factory site under controlled conditions; scalability to multi-facility and multi-shift environments remains unverified by any independent source. | 中 | SR019 |
| CR023 | NIST promotes a non-regulatory risk-based approach to AI measurement science, standards, and guidelines, with voluntary — not mandatory — adoption. | 中 | SR007 |
| CR024 | Production figures in the Chinese humanoid robot industry are not audited and cannot be independently verified, as confirmed by Humanoids Daily's reporting on the Unitree-Agibot shipment dispute. | 中 | SR013 |
| CR025 | Agibot and Unitree publicly disputed 2025 humanoid shipment leadership, with Agibot claiming 5,168 units and Unitree claiming 5,500, with no independent audit to resolve the discrepancy. | 中 | SR013 |
| CR026 | Over 150 humanoid robot companies currently operate in China as of early 2026, creating extreme competitive density in the sector. | 中 | SR010 |
| CR027 | China's NDRC warned against 'highly repetitive products' and 'speed and bubbles' in the humanoid robot sector and signalled a shift from broad subsidisation to selective support. | 中 | SR010 |
| CR028 | Unitree filed an IPO on the SSE STAR Market in March 2026, raising 4.202 billion RMB (~$580 million), providing the first revenue-disclosed Chinese humanoid company at public market scale. | 中 | SR014 |
| CR029 | Unitree's 2025 revenue exceeded 1.7 billion RMB ($235 million), up 335% year-over-year, with net profit exceeding 600 million RMB ($83 million), up 674%. | 中 | SR014 |
| CR030 | 92% of Chinese enterprise buyers in the Morgan Stanley survey require humanoid robot unit cost to fall below 200,000 yuan ($28,000 USD) before widespread adoption is feasible. | 中 | SR012 |
| CR031 | UBTECH's 2025 order book exceeded 800 million yuan ($113 million), with automotive and logistics as primary buyer segments — competing directly for Agilink's target customers. | 中 | SR027 |
| CR032 | Goldman Sachs' 2026 China survey showed the industry pivoting from 'universal imagination' to 'dedicated landing' in narrow use cases; general-purpose dexterous manipulation remains 1-3 years from commercial viability. | 中 | SR026 |
| CR033 | Chinese vendors accounted for 95% of 2025 global humanoid robot shipments, per IDC, implying that domestic market saturation may limit export-market differentiation. | 中 | SR015 |
| CR034 | Agibot retained approximately 80% of Agilink at incorporation on January 14, 2026, making Agilink operationally and financially dependent on its parent company. | 中 | SR021, SR028 |
| CR035 | Agibot chairman and CEO Deng Taihua was named as Agilink's legal representative at incorporation, indicating that Agilink's legal governance was tied to the Agibot executive at formation. | 中 | SR021 |
| CR036 | Agilink raised four funding rounds in under five months following its January 2026 incorporation, a pace Crunchbase described as part of the May 2026 robotics unicorn cohort. | 中 | SR028, SR029 |
| CR037 | Agibot's investor base includes BYD, SAIC Motor, BAIC Group, JD.com, and Tencent Holdings — state-adjacent enterprises whose involvement creates CFIUS-analogous scrutiny for non-Chinese secondary transactions. | 中 | SR017, SR020 |
| CR038 | Agilink's strategic investor base includes SAIC Financial Holdings and Joyson Electronics, both publicly listed Chinese industrial companies with strategic rationale to secure preferential supply relationships. | 中 | SR021 |
| CR039 | NDRC's 2026 shift toward selectivity in humanoid subsidies increases Agilink's policy concentration risk, as newly spun-out component companies may rank lower than established full-system suppliers in state support. | 低 | SR010 |
| CR040 | IFR noted that traditional industrial robots remain faster and more reliable for repetitive precision tasks than humanoids, suggesting that the addressable market for dexterous hands in conventional automation may be narrower than total humanoid market sizing implies. | 中 | SR018 |
| CR041 | Goldman Sachs identifies 15-20% annual production cost reduction and 20-hour battery life improvement as necessary thresholds for humanoid robots to reach viable unit economics in factory settings. | 中 | SR026 |
| CR042 | Agilink reported cumulative deliveries of more than 8,000 dexterous hands and 10,000 grippers as of the APC 2026 event in April 2026. | 低 | SR023 |
| CR043 | Agilink's OmniHand 2025 was priced at 9,800 RMB (~$1,350 USD), representing a 10× or greater price reduction versus predecessor research-grade products at $10,000-$200,000 USD. | 中 | SR023 |
| CR044 | Crunchbase confirmed Agilink reached a $1 billion valuation in May 2026 after four funding rounds since its January 2026 spinout from Agibot. | 中 | SR028 |
| CR045 | The IEEE Spectrum noted that collaborative robots took seven to ten years from first commercial versions to batch sales, suggesting dexterous hand products face a similarly extended commercial adoption curve. | 中 | SR016 |
| CR046 | No reviewed source disclosed any supply chain disruption incidents, component shortages, or product recalls affecting OmniHand production as of June 2026. | 低 | SR023, SR025 |
| CV001 | Agilink reached a valuation above $1 billion in May 2026 after completing its fourth funding round since incorporating on 2026-01-14, per Crunchbase News dated June 1, 2026. | 中 | SV001 |
| CV002 | Agilink raised four distinct funding rounds between January and May 2026, an exceptional five-month cadence to unicorn status in the dexterous hand sector. | 中 | SV001 |
| CV003 | Agilink's stated business model is dexterous hand technology with a licensing capability to the broader robotics market, as described by Crunchbase News. | 中 | SV001 |
| CV004 | The primary Crunchbase source for the $1 billion valuation does not specify pre-money or post-money convention, nor the size of the fourth round. | 中 | SV001 |
| CV005 | At a $1 billion valuation mark and an estimated cumulative revenue of $10–27 million USD (shipment proxy, 8,000+ units at 9,800–30,000 RMB blended ASP), the implied cumulative revenue multiple is approximately 37–100x (estimated, not audited). | 低 | SV001 |
| CV006 | Figure AI reached a post-money valuation of $39 billion in its September 2025 Series C funding round led by Parkway Venture Capital, with more than $2 billion in total capital raised since 2022. | 高 | SV002, SV010 |
| CV007 | Figure AI raised $675 million in its Series B in February 2024 at a $2.6 billion valuation with investors including Microsoft, OpenAI Startup Fund, NVIDIA, Jeff Bezos, and ARK Invest. | 高 | SV003, SV002 |
| CV008 | Figure AI's valuation increased approximately 15x in 18 months, from $2.6 billion (Series B, February 2024) to $39 billion (Series C, September 2025), a step-up with no revenue-growth anchor. | 中 | SV002, SV003 |
| CV009 | Apptronik closed over $935 million in total Series A funding by February 2026 at a reported $5 billion valuation, with investors including Google DeepMind, Mercedes-Benz, John Deere, and the Qatar Investment Authority. | 高 | SV004, SV005, SV013 |
| CV010 | A US SEC Form D filing by HII Apptronik Series I (filed November 2025) confirms the existence of an institutional investment special purpose vehicle associated with Apptronik's Series A extension. | 高 | SV016, SV004 |
| CV011 | Agility Robotics reached approximately $1.75 billion post-money valuation after its Series C in early 2025, with Amazon and GXO as deployment partners. | 中 | SV012 |
| CV012 | 1X Technologies raised $100 million in its Series B in January 2024 led by the OpenAI Startup Fund; implied valuation is analyst-estimated at $0.5–1 billion and is not publicly confirmed. | 中 | SV014, SV015 |
| CV013 | Unitree disclosed 2025 revenue exceeding 1.7 billion RMB ($235 million), up 335% year-over-year, with net profit exceeding 600 million RMB ($83 million), in its SSE STAR Market IPO application filed March 2026. | 高 | SV007, SV027 |
| CV014 | Unitree is the only Chinese humanoid robotics company with publicly disclosed audited revenue data as of June 2026, making it the only verifiable sector floor for unit economics. | 高 | SV007, SV027 |
| CV015 | Agibot, Agilink's 80%-shareholder parent, exceeded a CNY 7 billion (~$960 million) pre-money valuation at its August 2025 LG Electronics and Mirae Asset strategic investment round. | 中 | SV008, SV009 |
| CV016 | Global robotics startups raised nearly $14 billion in funding in 2025, up from $8.2 billion in 2024, with China-based robotics companies raising $5.6 billion through mid-May 2026 alone. | 高 | SV006, SV019 |
| CV017 | Skild AI tripled its valuation to above $14 billion following a $1.4 billion raise in January 2026, illustrating the AI-platform multiple premium available to robotics software/AI companies versus pure hardware suppliers. | 中 | SV006 |
| CV018 | China's 15th Five-Year Plan (2026–2030) makes AI-powered robotics a core national strategy, with China's domestic supplier share rising from 30% in 2020 to 57% in 2024 and over 50% of global annual industrial robot installs deployed in China. | 高 | SV019, SV006 |
| CV019 | Hardware startup valuations at Agilink's pre-revenue stage are typically driven by strategic optionality and comparable benchmarking rather than revenue multiples, as conventional P/S methodology requires disclosed revenue to apply. | 中 | SV017, SV030 |
| CV020 | Goldman Sachs identifies data flywheel, manufacturing scale, and industrial deployment proof as the primary value drivers for private humanoid robotics valuations in the 2025–2026 cohort. | 高 | SV017, SV023 |
| CV021 | Morgan Stanley's humanoid robotics research identifies component suppliers (citing Inovance and Leaderdrive as analogues) as among the earliest beneficiaries of humanoid robot supply chain build-out, validating a component-tier investment thesis. | 中 | SV022 |
| CV022 | IDC forecasts global humanoid robot shipments will exceed 510,000 units by 2030 at approximately 95% CAGR, with Chinese vendors accounting for 95% of 2025 shipments. | 高 | SV018, SV017 |
| CV023 | Agilink's OmniHand 2025 is priced at 9,800 RMB (~$1,350 USD), representing a 7–150x price reduction versus predecessor research-grade dexterous hands, positioning it as the market-access entry point. | 中 | SV001 |
| CV024 | Agilink reported shipping more than 8,000 dexterous hands and 10,000 grippers cumulatively by May 2026, as reported by Pandaily and other sector media; these figures are company-reported and not independently audited. | 中 | SV001 |
| CV025 | A shipment-based revenue proxy at OmniHand 2025 list price (9,800 RMB × 8,000 units) implies cumulative receipts of approximately 78 million RMB (~$10.8 million USD); at a blended 30,000 RMB ASP the estimate reaches approximately $33 million USD — both are speculative estimates with significant uncertainty. | 低 | SV001 |
| CV026 | A conventional hardware P/S multiple of 5–15x forward revenue would require Agilink to demonstrate annual revenue of $67–200 million USD to support a $1 billion valuation — a level not achievable from current disclosed shipment data on any plausible pricing assumption. | 低 | SV017, SV030 |
| CV027 | As of June 2026, Agilink has made no public disclosure of revenue, gross margin, operating expenses, burn rate, or any other financial metric in any jurisdiction. | 中 | SV001 |
| CV028 | Agilink's Chinese origin creates a structural geopolitical discount relative to equivalent Western-market valuations, reflecting the ASRA legislative risk, BIS export controls, and CFIUS/enterprise procurement chilling effects. | 中 | SV021, SV019 |
| CV029 | Agibot retains approximately 80% of Agilink at incorporation with Deng Taihua as legal representative; no independent board, preference terms, or minority protection covenants have been publicly disclosed. | 中 | SV001, SV025 |
| CV030 | Component-supplier companies in hardware ecosystems typically trade at 5–15x forward revenue multiples, versus 20–50x multiples available to AI platform or robotics foundation-model companies with recurring licensing revenue. | 中 | SV017, SV022 |
| CV031 | A proprietary AI manipulation dataset and licensed dexterous hand model sold to third-party OEMs would shift Agilink's applicable valuation multiple from hardware (5–15x) toward AI platform (20–50x forward revenue), consistent with Skild AI's tripled valuation on a similar platform thesis. | 低 | SV006, SV017 |
| CV032 | Chinese robotics companies historically trade at a 30–50% discount to US VC marks on equivalent fundamentals when STAR Market price-earnings multiples are applied, based on Unitree IPO context and analyst estimates of Chinese humanoid sector pricing. | 低 | SV007, SV027 |
| CV033 | Tianji, a Chinese robot-arm component supplier, raised $147 million in its Series B in May 2026 at a $1.5 billion valuation led by Hillhouse Capital and Meituan, with a backlog exceeding 10,000 units from 45 OEM customers. | 中 | SV001, SV006 |
| CV034 | China's NDRC explicitly warned in 2026 against 'highly repetitive products' and 'speed and bubbles' in the humanoid sector, and noted that over 150 humanoid robot companies are now operating in China. | 高 | SV021, SV024 |
| CV035 | Production figures in the Chinese humanoid robotics industry are not independently audited and cannot be externally verified; all shipment counts — including Agilink's 8,000+ unit claim — are company-reported. | 高 | SV020, SV027 |
| CV036 | Unitree and Agibot disputed their respective 2025 shipment counts (Unitree claiming 5,500; Agibot claiming 5,168), with Humanoids Daily noting that 'what counts as a humanoid' is itself contested between bipedal and wheeled designs. | 高 | SV020, SV027 |
| CV037 | Figure AI's $39 billion September 2025 valuation is characterised by analysts as a single-investor mark (Parkway Venture Capital) with limited secondary market participation and secondary-market volatility reported by TechCrunch. | 中 | SV002, SV010 |
| CV038 | No prior-round benchmark exists for a pure-play dexterous hand component company at $1 billion valuation; Agilink's mark is without direct comparables, making the valuation difficult to challenge or confirm against sector precedent. | 中 | SV001, SV006 |
| CV039 | In a bull scenario, Agilink secures 10 or more OEM licensees and reaches annual deliveries of 100,000+ units by 2027 with an AI-model licensing stream, supporting a $3–8 billion valuation by 2028 on 20–30x forward revenue multiples (estimated). | 低 | SV017, SV018 |
| CV040 | In a base scenario, Agilink reaches 20,000–50,000 units per year by 2027 with Agibot as the dominant channel, sustaining its $800 million to $1.5 billion valuation range on hardware multiples consistent with Agility Robotics' comparable (estimated). | 低 | SV012, SV001 |
| CV041 | In a bear scenario, humanoid robotics sector multiple compression forces Agilink to a $100–400 million valuation range, particularly if Agibot reintegrates the supply chain, the sim-to-real gap stalls commercialisation, or a down-round is triggered by capital constraints (estimated). | 低 | SV021, SV020 |
| CV042 | The strategic premium for end-effector component suppliers only unlocks at scale if Agilink establishes cross-OEM supply relationships that no individual full-system player can replicate through in-house manufacturing; below that threshold, the multiple compresses toward hardware norms. | 中 | SV017, SV022 |
| 编号 | 出版方 | 标题 | 引文 |
|---|---|---|---|
| SO001 | Pedaily | 智元旗下灵巧手公司「临界点 AGILINK」连续完成两轮融资 | |
| SO002 | Tencent News | 头部互联网大厂领投,「临界点」再获数亿元融资 | |
| SO003 | Tencent News | 估值突破10亿美元!临界点完成新一轮数亿元融资 | |
| SO004 | Sina Finance | 智元拆出一家“灵巧手”独角兽,成立不足5月估值破10亿美元 | |
| SO005 | Huxiu | AGILINK完成新一轮数亿元融资,专注具身智能灵巧手 | |
| SO006 | NetEase | 临界点完成数亿元融资,成立不足5月估值破10亿美元 | |
| SO007 | Economic Observer | 临界点完成新一轮数亿元人民币融资,公司估值突破10亿美元 | |
| SO008 | Gasgoo | Seeds | AGILINK Raises Hundreds of Millions | |
| SO009 | Humanoids Daily | AGILINK Hits Unicorn Status: AGIBOT’s Dexterous Hand Spin-off Raises Hundreds of Millions | |
| SO010 | 36Kr Europe | "Critical Point" Secures Hundreds of Millions in New Financing, Led by Leading Internet Giants | |
| SO011 | Sohu | 头部互联网大厂领投,「临界点」再获数亿元融资丨智能涌现独家 | |
| SO012 | Zhihu | 智元成立“临界点”,国内30家灵巧手公司深度行研(下) | |
| SO013 | AGIBOT | AGIBOT official homepage | |
| SO014 | AGIBOT | About Us | |
| SO015 | AGIBOT | AGIBOT Unveils New Generation of Embodied AI Robots and Models at APC 2026 | |
| SO016 | Pandaily | AGILINK Launches OmniHand 3 Series, Ships Over 8,000 Dexterous Hands | |
| SO017 | Crypto Briefing | AGILINK raises hundreds of millions at $1B valuation, shipping thousands of robotic hands since January | |
| SO018 | AGIBOT | OmniHand O10 product page | |
| SO019 | AGIBOT | OmniHand O12 / OmniHand Pro product page | |
| SO020 | Humanoid.guide | OmniHand Pro 2025 – Advanced Industrial Robotic Hand | |
| SO021 | GitHub | AgibotTech/Omnihand-2025-SDK | |
| SO022 | China Youth International | Agibot Releases OmniHand 2025 Dexterous Hand Series | |
| SO023 | Botonly | AGIBOT OmniHand 2025 Robotic Hand | Specs & Buy | |
| SO024 | Robots Canada | AgiBot OmniHand Series | Dexterous Robot Hand (2025) | |
| SO025 | Humanoids Daily | Deployment Year One: AGIBOT Unveils Massive Fleet and AI Model Stack at APC 2026 | |
| SO026 | AGIBOT Store | Official marketplace entry for OmniHand Pro 2025 | |
| SO027 | Humanoids Daily | China’s Robot Fever Meets Reality: Beijing Warns of “Bubbles” and “Repetitive” Clones | |
| SM001 | Goldman Sachs Research | Humanoid Robot: The AI Accelerant | global market for humanoid robots may reach US$6bn in 10-15 years … up to US$154bn by 2035E in a blue-sky scenario |
| SM002 | Goldman Sachs | Humanoid Robots: Sooner Than You Might Think | humanoid robot solutions could be economically viable in factory settings between 2025 to 2028 |
| SM003 | MarketsandMarkets | Humanoid Robot Market — Global Forecast to 2030 | global humanoid robot market is expected to grow from USD 2.92 billion in 2025 to USD 15.26 billion in 2030, at a CAGR of 39.2% |
| SM004 | IDC | From Task Execution to Value Creation: What the 2026 Humanoid Robot Half Marathon Reveals About Industry Progress | global shipments of humanoid robots are expected to exceed 510,000 units by 2030, representing a CAGR of nearly 95% |
| SM005 | International Federation of Robotics | China Makes AI-powered Robots Core of National Strategy | China's manufacturing industry already has an operational stock of around 2 million units — approximately 4.5 times more than the global no. 2, Japan. 54% of annual industrial robots installed worldwide were deployed in China. |
| SM006 | Robotics and Automation News | Why China's New Humanoid Robot Standards Could Change the Industry | nearly 80 percent of tasks where humans excel but traditional automation struggles are strongly related to tactile sensing |
| SM007 | Taibo Financial (translating Goldman Sachs China survey) | Goldman Sachs' Research on Humanoid Robots in China: The Industry is Shifting from Universal Imagination to Dedicated Landing | Simulate to Reality gap still a bottleneck; accuracy of 80-90% in simulated environments often drops below 50% in real scenarios |
| SM008 | Grand View Research | Humanoid Robot Market Size, Share & Trends Analysis Report | |
| SM009 | Forbes | Humanoid Robots: Here Are The 16 Leading Manufacturers | Goldman Sachs says selling humanoid robots will be a $38 billion space by 2035, while Ark Invest says the resulting economic value of their labor could be as high as $24 trillion |
| SM010 | IEEE Spectrum | Humanoid Robots Are Getting to Work | Collaborative robots took roughly seven to 10 years from their first commercially available versions to batch sales |
| SM011 | Humanoids Daily (reporting Morgan Stanley AlphaWise survey) | Morgan Stanley Survey: Chinese Companies Want Humanoids, But the Products Aren't Ready | only 23% of those surveyed expressed satisfaction with current products, citing issues with dexterity, functionality, and price |
| SM012 | Humanoids Daily | China's Robot Fever Meets Reality: Beijing Warns of 'Bubbles' and 'Repetitive' Clones | over 150 humanoid robot companies now operating in China; speed and bubbles are perennial issues in frontier industries |
| SM013 | Crunchbase News | Amid Record Robotics Funding, Apptronik Raises $520M Series A Extension | Startups in the sector raised nearly $14 billion in funding in 2025, up from $8.2 billion in 2024 |
| SM014 | Humanoids Daily | UBTECH Claims 'World's First' Mass Delivery of Humanoids as 'Robot Army' Deploys | 2025 order book has now exceeded 800 million yuan; industrial client list includes BYD, Geely Auto, FAW-Volkswagen, Dongfeng Liuzhou Motor, and Foxconn |
| SM015 | Humanoids Daily | China Drafts 'Dream Team' for Humanoid Robot Standards, Taps Founders of Unitree and AgiBot to Lead | Wang Xingxing, founder of Unitree Robotics, and Peng Zhihui, the co-founder of AgiBot, have been appointed as deputy chairs of the committee |
| SM016 | Yicai Global | Mirae Asset-LG Electronics Fund Invests in Chinese Humanoid Robot Startup AgiBot | AgiBot has completed 10 financing rounds, with investment from big firms such as Tencent Holdings, JD.Com, BYD, SAIC Motor, BAIC Group, and TCL |
| SM017 | Humanoids Daily | The Great Numbers War: Unitree Claims Global Shipment Lead, Rebuts AgiBot's 'No. 1' Title | production figures in this industry are not audited and cannot be independently verified |
| SM018 | US News & World Report (Reuters) | US Lawmakers to Introduce Bill to Ban Government Use of Chinese Robots | Two U.S. senators planned to introduce a bill that would ban the government from buying or operating humanoid robots made by Chinese firms |
| SM019 | Robotics and Automation News | Agibot Secures Strategic Investment from LG Electronics and Mirae Asset | |
| SM020 | Humanoids Daily | Unitree Files for $580M IPO: Humanoid Sales Surpass Robot Dogs as Profits Soar | 2025 Revenue: Over 1.7 billion RMB (approx. $235 million USD), a 335% increase year-over-year |
| SM021 | Interesting Engineering | AGIBOT Puts Humanoid Robots to Work in Real Factory Production | throughput of up to 310 units per hour, cycle times of around 19 to 20 seconds per task, and a success rate exceeding 99.9 percent |
| SM022 | Crunchbase News | AI Services And Robotics Lead Diverse Crop Of 29 New May Unicorns | Shanghai-based Agilink has raised four funding rounds since it was spun out of Agibot in January, and reached a valuation of $1 billion. Agilink is focused on dexterous hand technology. |
| SM023 | Humanoids Daily | Agilink Hits Unicorn Status: Agibot's Dexterous Hand Spin-Off Raises Hundreds of Millions | |
| SM024 | Pandaily | Agibot Completes New Funding Round, Valuation Exceeds 7 Billion Yuan | |
| SM025 | Humanoids Daily | Deployment Year One: Agibot Unveils Massive Fleet and AI Model Stack at APC 2026 | |
| SP001 | Humanoid Index | Shadow Robot Company: Funding, Valuation, Robot Specs & More | Dexterous Hand. World-leading robot hand with 24 degrees of freedom. Research standard. |
| SP002 | Humanoid Index | Shadow Dexterous Hand by Shadow Robot Company: Specs, Price & Status | 8/10 Humanoid Index Assessment. Use Cases: Research. |
| SP003 | arXiv (CMU / RSS 2023) | LEAP Hand: Low-Cost, Efficient, and Anthropomorphic Hand for Robot Learning | LEAP Hand is low-cost and can be assembled in 4 hours at a cost of 2000 USD from readily available parts. LEAP Hand significantly outperforms its closest competitor Allegro Hand in all our experiments while being 1/8th of the cost. |
| SP004 | GitHub (leap-hand org) | LEAP Hand API: API for controlling LEAP Hand v1 | LEAP Hand: Low-Cost, Efficient, and Anthropomorphic Hand for Robot Learning |
| SP005 | Humanoid.guide | Allegro Hand V4 – Adaptive Robotic Hand by Wonik Robotics | Degrees of freedom, hands: 16. Strength [kg]: 5. Price: $24,500. Tactile/Force Sensors: No. |
| SP006 | Humanoid.guide | Robotera XHAND1 – 12-DoF Dexterous Robotic Hand | 12 active DoF, rich tactile sensing, and strong gripping force. Strength [kg]: 25. Price: $14,000. Tactile/Force Sensors: Yes. |
| SP007 | Humanoid Index | GR-2 by Fourier Intelligence: Specs, Price & Status | GR-2 features an industry-leading 53 degrees of freedom. DOF (Hands): 12 per hand. Six array-type tactile sensors. Price: ~$100,000+. |
| SP008 | Humanoid Index | Fourier Intelligence: Funding, Valuation, Robot Specs & More | SoftBank-backed. Strong healthcare rehabilitation roots. Stage: Series C Plus. Employees 800+. Last Round: Series D (2024) Verified. |
| SP009 | The Robot Report | RightHand Robotics reveals RightPick 3 item-handling robot | Pick rate of up to 1,200 units per hour. Level 4 autonomy with greater than 99% pick accuracy. RightHand Robotics closed a $23 million Series B round in December 2018. |
| SP010 | RightHand Robotics | RightHand Robotics — RightPick System Overview | The best-in-class piece-picking solution for reliable and autonomous order fulfillment. |
| SP011 | Dexterity AI | Dexterity Blog — Physical AI, Foresight, Instinct | Trained with experience from over 100 million autonomous actions in production, it powers safety-rated Physical AI in Fortune 50 enterprises. |
| SP012 | Inspire Robots (Beijing Yinshi) | Inspire Robots — Official Homepage (Chinese) | 北京因时机器人科技有限公司创立于2016年. RH56BFX, RH56DFX, RH56E2, RH56F1, RH5DG2, RH56H1 series listed. |
| SP013 | Fourier Intelligence (FFTAI) | Fourier Intelligence — Official Website | At Fourier, our mission is to leverage full-stack robotics technology to enrich people's lives. |
| SP014 | Unitree Robotics | Unitree R1 Official Product Page | Price from $4,900. 20-26-DOF for adaptation to complex scenarios. To adapt to diverse application scenarios, the robot's end effectors can be replaced with a variety of different actuators. |
| SP015 | CnTechPost | Unitree releases dual-arm humanoid robot R1 | Chinese robot maker Unitree released its new R1 series dual-arm humanoid robot on Thursday, with a starting price set at 26,900 yuan ($3,930). The company has currently showcased dexterous hands with two, three, and five fingers. The humanoid robot business transitioned into its largest source of revenue in 2025, driving a massive 335.4% surge in the company's annual revenue last year. |
| SP016 | RoboticsTomorrow | ROBOTERA Raises Over USD 200 Million in New Round Led by SF Group, HSG and IDG Capital | ROBOTERA has raised over USD 200 million in a new financing round led by SF Group. A key focus is its pioneering full direct-drive dexterous hand architecture, the first of its kind in the industry. Its hardware system has been adopted by leading global technology companies and research institutions, including Boston Dynamics, NVIDIA, and Apple. |
| SP017 | KrASIA | More than a hardware play: Robot Era's founder sets the record straight on the company's true ambitions | Some people think we're just a company that builds dexterous hands. It counts nine of the world's ten most valuable tech companies among its clients. |
| SP018 | CnTechPost | Embodied AI startup Robotera raises $200 million new funding round | The funding comes less than a month after the company completed a 1 billion yuan ($146 million) round. Robotera plans to begin mass delivery of thousands of robots in the second quarter of 2026. |
| SP019 | Agibot (official) | AGIBOT APC 2026 — OmniHand 3 Ultra-T Announcement | OmniHand 3 Ultra-T features a 22+3 DOF tendon-driven system, a lightweight 500g design, and a 10:1 load-to-weight ratio. With full-hand 3D tactile sensing, an integrated palm camera, sub-0.3s response time. |
| SP020 | Humanoid.guide | OmniHand by AgiBot — Product Page | Price: $3,775. Degrees of freedom, hands: 16. Strength [kg]: 2. Weight [kg]: 0.5. Tactile/Force Sensors: Yes. |
| SP021 | Humanoid.guide | OmniHand Pro by AgiBot — Product Page | Price: $12,700. Degrees of freedom, hands: 19. Strength [kg]: 35. Weight [kg]: 0.75. Tactile/Force Sensors: Yes. |
| SP022 | Agibot (Genie Studio) | Genie Studio — Developer Platform for Embodied AI | One-Click Deployment: Seamless algorithm migration from cloud to real devices, achieving 2-3x performance gains over traditional single-GPU inference solutions. |
| SP023 | AimRT Project (Agibot open-source) | AimRT — Open-Source Robot Middleware Release Notes | AimRT v1.7.0 Release. Added ROS 2 Jazzy support. Zenoh plugin, gRPC, MQTT, MCAP recording, topic logger. |
| SP024 | Aparo Robot | Year One of Deployment: How AgiBot Is Building a Global Robot Empire | AgiBot's 10,000th unit. AgiBot partner conference in April 2026. Partners from 17 countries. |
| SP025 | Humanoids Daily | The Great Numbers War: Unitree Claims Global Shipment Lead, Rebuts Agibot's No.1 Title | Unitree claims global shipment lead, rebuts Agibot's No.1 title — the contest over who leads in humanoid shipments casts doubt on Agibot/Agilink's scale claims. |
| SI001 | South China Morning Post | Unicorn born in record time amid 'arms race' among China's robotic hand developers | "Achieving unicorn status in less than 150 days is unprecedented in the humanoid component sector," said Wu Meimei, senior analyst at ITJuzi, which tracks China's venture-capital market. |
| SI002 | South China Morning Post | China's top economic planning agency warns of rise of too many robots | "Speed and bubbles have always been issues that need to be grasped and balanced in the development of cutting-edge industries," Li Chao, NDRC spokesperson, said of China's humanoid robot sector. |
| SI003 | MarketScreener (S&P Capital IQ data) | Shanghai Agilink Innovation Intelligent Technology Co., Ltd. announced that it has received funding from a group of investors | Investors in Round 3 include Baidu Venture, Yunfeng Fund LP, Hangzhou Longqi Scientific Investment, Nantong Wofu Jinxin, Beijing MOOC-CN Jinxin, Ningbo Joyson Electronic Corp., SAIC Motor Financial Holdings, C Capital, BRV Partners, Shanghai Hillhouse Equity Investment Management, and Hillhouse Investment Management. |
| SI004 | C Capital | C Capital Invests in AGILINK's Multi-Hundred-Million RMB Financing Round | "We're excited to announce that C Capital has invested in AGILINK's newly announced multi-hundred-million RMB financing round." |
| SI005 | AiFlashDesk | 临界点AGILINK完成数亿元融资:前智元机器人团队领衔,深耕具身智能灵巧手 | Xiong Kun categorizes the dexterous hand market into three tiers: entry (6–10 DoF, few thousand to 30,000 RMB), mid-range (11–17 DoF, 50,000–100,000+ RMB), and high-end (20+ DoF, 200,000+ RMB). |
| SI006 | Unite.AI | China Warns of Bubble Risk as 150 Companies Flood Humanoid Robot Market | "Frontier industries have long grappled with the challenge of balancing the speed of growth against the risk of bubbles — an issue now confronting the humanoid robot sector as well," Li Chao said, noting that "humanoid robots have yet to fully mature in terms of technological pathways, business models and application scenarios." |
| SI007 | Financial Times Markets (PR Newswire) | Hyperscale Data's Subsidiary Omnipresent Robotics Enters into an Agreement Providing for the Acquisition of Robots from AGIBOT and Related Developments | "AGIBOT agreed to sell up to 143 intelligent robot products to Omnipresent, authorize Omnipresent to resell such products under Omnipresent's brand, and assist Omnipresent in establishing a robotics data collection center at Hyperscale Data's Michigan Data Center." |
| SI008 | PR Newswire | Hyperscale Data's Subsidiary Omnipresent Robotics Enters into an Agreement Providing for the Acquisition of Robots from AGIBOT and Related Developments | |
| SI009 | CSRC / Joyson Electronics via cninfo.com.cn | 宁波均胜电子股份有限公司 2025年年度报告摘要 (Ningbo Joyson Electronic Corp. 2025 Annual Report Summary) | Joyson 2025 annual report discloses strategic expansion into the "新兴智能体产业链" (emerging intelligent body supply chain), consistent with its confirmed investment in Agilink's Round 3. |
| SI010 | Reuters | China's Agibot spinoff Agilink valued at $1 billion in latest funding round | |
| SI011 | Humanoids Daily | AGILINK Hits Unicorn Status: AGIBOT's Dexterous Hand Spin-off Raises Hundreds of Millions | |
| SI012 | Humanoids Daily | Deployment Year One: Agibot Unveils Massive Fleet and AI Model Stack at APC 2026 | |
| SI013 | Pandaily | AGILINK Launches OmniHand 3 Series, Ships Over 8,000 Dexterous Hands | |
| SI014 | Robots Canada | AgiBot OmniHand Series: Complete Guide to AgiBot's Dexterous Robot Hand Lineup | "At 9,800 RMB, the OmniHand 2025 was announced as the first high-DOF dexterous hand priced under 10,000 RMB, substantially broadening access to dexterous manipulation hardware... This pricing represents a substantial reduction from previous commercial dexterous hands, which typically started at $10,000 and could reach $200,000 for research-grade platforms." |
| SI015 | Botonly | AGIBOT OmniHand 2025 Robotic Hand — Price, Specs & Buying Guide | Indicative pricing: From $4,420 USD (Last updated June 1, 2026) |
| SI016 | AGIBOT (official) | AGIBOT Unveils New Generation of Embodied AI Robots and Models, Accelerating Real-World Deployment of Physical AI | "OmniHand 3 Ultra-T represents the next-generation upgrade of the OmniHand portfolio, delivering industry-leading, human-level dexterity. It features a 22+3 DOF tendon-driven system, a lightweight 500g design, and a 10:1 load-to-weight ratio." |
| SI017 | AGIBOT (official) | OmniHand — AGIBOT Product Page | |
| SI018 | Gasgoo (Seeds section) | Seeds | AGILINK Raises Hundreds of Millions | Gasgoo | |
| SI019 | Humanoids Daily | China's Robot Fever Meets Reality: Beijing Warns of 'Bubbles' and 'Repetitive' Clones | "The map of China's robotics industry is becoming incredibly crowded, and Beijing has decided it is time to thin the herd." |
| SI020 | Pedaily | Agilink completes two early-stage financing rounds led by Hillhouse and BlueRun | |
| SI021 | Economic Observer (EEO) | Agilink unicorn financing confirmed — EEO report | |
| SI022 | Sina Finance | Agilink dexterous hand spin-off unicorn status confirmed | |
| SI023 | QQ News (Tencent) | Agilink multi-hundred-million yuan February round announcement | |
| SI024 | Humanoids Daily | Unitree Files for $580M IPO; Humanoid Sales Surpass Robot Dogs as Profits Soar | |
| SI025 | Humanoids Daily | China Drafts Dream Team for Humanoid Robot Standards; Taps Founders of Unitree and Agibot to Lead | |
| SI026 | Crunchbase News | New Unicorn Startups — May 2026: OpenAI, Anthropic IPOs, SpaceX, Robotics | |
| SI027 | Huxiu | Agilink early-stage coverage and embodied AI analysis | |
| SE001 | Agibot (official) | OmniHand 2025 product specifications page (O10) | Active DOF: 10; Total DOF: 16; Transmission: Linkage mechanism; Drive: Motor + Gear; 400+ tactile points; Sensing Range: 0~20N |
| SE002 | Agibot (official) | OmniHand Pro 2025 product specifications page (O12) | Active DOF: 12; Total DOF: 19; Transmission: Linkage transmission; Drive Method: Motor + Screw Rod; Load capacity (palm up, 5 cm rigid object): 35kg; Finger reposition accuracy (typ.): 0.02mm |
| SE003 | Agibot (official) | AGIBOT Unveils New Generation of Embodied AI Robots — APC 2026 Product Announcement | OmniHand 3 Ultra-T: 22+3 DOF tendon-driven system, a lightweight 500g design, and a 10:1 load-to-weight ratio. With full-hand 3D tactile sensing, an integrated palm camera, sub-0.3s response time, and a wide wrist range (55° pitch, 40° yaw) |
| SE004 | AgibotTech (GitHub) | Omnihand-2025-SDK — Official OmniHand SDK repository | Operating System: Ubuntu 22.04 (x86_64); Python: 3.10 or higher; Build Tool: CMake 3.24 or higher; CANFD: Currently supports ZLG USBCANFD series |
| SE005 | Robots Canada | AgiBot OmniHand Series: Complete Guide to AgiBot's Dexterous Robot Hand Lineup | At 9,800 RMB, the OmniHand 2025 was announced as the first high-DOF dexterous hand priced under 10,000 RMB, substantially broadening access to dexterous manipulation hardware |
| SE006 | Humanoids Daily | Deployment Year One: AGIBOT Unveils Massive Fleet and AI Model Stack at APC 2026 | AGIBOT declared 2026 as 'Deployment Year One,' signaling a strategic pivot from showcasing technical capabilities to delivering measurable ROI in real-world workflows. |
| SE007 | Botonly | AGIBOT OmniHand 2025 Robotic Hand — Price, Specs & Buying Guide | Indicative pricing: From $4,420 USD. Last updated June 1, 2026 |
| SE008 | Humanoid Guide | OmniHand Pro 2025 — Advanced Industrial Robotic Hand | $12 700 — OmniHand Pro is an advanced robotic hand with 19 degrees of freedom, multi-modal tactile sensing, and human-like dexterity for structured manipulation and industrial tasks. |
| SE009 | China Youth International | Agibot Releases OmniHand 2025 Dexterous Hand Series | The Agile Edition OmniHand 2025 is available at a limited-time discount of 5,000 yuan, priced at 9,800 yuan — marking the first time a high-DOF dexterous hand has entered the sub-10,000 yuan range |
| SE010 | Agibot (Genie Studio) | Genie Studio — Full-Stack Embodied AI Development Platform | Supports asynchronous data collection from various robot bodies and end effectors; Supports RDT, Pi0, OpenVLA, GR00T, and proprietary AgiBot Go-1 foundation models; scenario adaptation in 48 GPU-days |
| SE011 | AimRT Project | AimRT — Open Source Robot Middleware | |
| SE012 | Agibot Store (official) | Agibot Store — OmniHand product listing | |
| SE013 | Humanoid Guide | Robotera XHAND1 — 12-DoF Dexterous Robotic Hand | XHAND1 — 12 active DoF, $14,000, 1.1 kg, 25 kg strength, gear-driven force-controlled joint modules |
| SE014 | Humanoid Guide | LEAP Hand by Carnegie Mellon University — Robotic Hand Specs | $2 000 — An affordable, research-focused robotic hand that combines dexterity, tactile sensing, and accessibility for advanced manipulation tasks. 16 DoF, 0.94 kg, servo motors, ~2 kg payload |
| SE015 | LEAP Hand Project (v1.leaphand.com) | LEAP Hand: Low-Cost, Efficient, and Anthropomorphic Hand for Robot Learning | LEAP Hand: Low-Cost, Efficient, and Anthropomorphic Hand for Robot Learning — Shaw, Agarwal, Pathak; Robotics: Science and Systems (RSS), 2023 |
| SE016 | Wonik Robotics (Allegro Hand official) | Allegro Hand — Product Overview (allegrohand.com) | Allegro Hand V4 — cost-effective and highly adaptive robotic hand; V5 Sense — 16 pressure sensing channels and human-like versatile grasping capability |
| SE017 | Wonik Robotics (Allegro Hand official) | Allegro Hand V4 Product Page | |
| SE018 | Wonik Robotics (Allegro Hand official) | Allegro Hand V4 Tech Info | |
| SE019 | HuggingFace (agibot-world) | AgiBot World — Large-Scale Robotic Learning Dataset | AgiBot World Beta dataset contains 1M+ trajectories, with a total duration of 2,976.4 hours, covering 217 specific tasks, 87 skills, 3,000+ different objects, and 100+ real-world scenarios |
| SE020 | Wikipedia | Robotic arm — Wikipedia | Anthropomorphic robot: shaped in a way that resembles a human hand, i.e. with independent fingers and thumbs. |
| SE021 | Humanoid Index | GR-2 by Fourier Intelligence — Specs & Status | GR-2 includes hands with 12 degrees of freedom each and six array-type tactile sensors |
| SE022 | Agibot (official) | AGIBOT Declares 2026 'Deployment Year One' at APC 2026 | AGIBOT announced the rollout of its 10,000th robot as of March 2026; seven standardized productivity solutions targeting high-value industry scenarios |
| SE023 | Pandaily | Agilink Launches OmniHand 3 Series, Ships Over 8,000 Dexterous Hands | |
| SE024 | Humanoid Guide | OmniHand 2025 Product Profile — Humanoid.guide | |
| SE025 | Humanoid Index | XHAND1 from Inspire Robots — Humanoid Index | |
| SU001 | PRNewswire / AGIBOT | AGIBOT and Longcheer Technology Achieve World's First Embodied AI Deployment in Consumer Electronics Precision Manufacturing Mass-Production Line | In just four months, the AGIBOT G2 was integrated into Longcheer's mass production line, delivering stable, continuous operation and meeting all key targets. |
| SU002 | CnTechPost | Agibot G2 achieves zero errors in 8-hour continuous factory operation | The Agibot G2 robot completed 8 hours of continuous operation at Longcheer's tablet factory, executing 2,283 tasks with zero errors. |
| SU003 | AGIBOT (official) | AGIBOT and Longcheer Technology Achieve World's First Embodied AI Deployment in Consumer Electronics Precision Manufacturing Mass-Production Line | |
| SU004 | AGIBOT (official) | How AGIBOT's Seven Solutions Are Reframing the Commercialization of Embodied AI | At Longcheer Technology, robots handle up to 5,700 units per day in continuous 24/7 operation. |
| SU005 | AGIBOT (official) | The First Hong Kong Embodied AI Industry Summit and AGIBOT Partner Conference 2026 Hong Kong Opens | |
| SU006 | PRNewswire / AGIBOT | AGIBOT Makes Its U.S. Market Debut at CES 2026 with Its Full Humanoid Robot Portfolio | 5,000 robots to date … already operational across eight core commercial applications |
| SU007 | Humanoids Daily | The 10,000-Unit Threshold: AGIBOT Accelerates Production in Bid for Global Dominance | |
| SU008 | Aparo Robot | Year One of Deployment: How AgiBot is Building a Global Robot Empire | In a major play for the Southeast Asian market, AgiBot signed a Memorandum of Understanding (MOU) with Genting Malaysia Bhd on April 17, 2026. |
| SU009 | Humanoid Index | AgiBot: Funding, Valuation, Robot Specs & More | Units Shipped: 5,100 (2025) Verified |
| SU010 | KrASIA | More than a hardware play: Robot Era's founder sets the record straight on the company's true ambitions | It counts nine of the world's ten most valuable tech companies among its clients, including Haier Smart Home, Lenovo, and BZS Technology Development. |
| SU011 | RoboticsTomorrow | ROBOTERA Raises Over USD 200 Million in New Round Led by SF Group, HSG and IDG Capital | deployments across more than ten logistics centers in collaboration with China Post and SF Group |
| SU012 | The Robot Report | RightHand Robotics reveals RightPick 3 item-handling robot | |
| SU013 | Longcheer Technology (official) | Longcheer Technology Corporate Website | |
| SU014 | Interesting Engineering | AGIBOT puts humanoid robots to work in real factory production | |
| SU015 | Humanoids Daily (citing Morgan Stanley) | Morgan Stanley Survey: Chinese Companies Want Humanoids, But the Products Aren't Ready | only 23% of those surveyed expressed satisfaction with current products, citing issues with dexterity, functionality, and price |
| SU016 | U.S. News & World Report (Reuters) | US Lawmakers to Introduce Bill to Ban Government Use of Chinese Robots | would ban the government from buying or operating humanoid robots made by Chinese firms |
| SU017 | Pandaily | AGILINK Launches OmniHand 3 Series, Ships Over 8,000 Dexterous Hands | |
| SU018 | AGIBOT (official) | AGIBOT Declares 2026 'Deployment Year One' at APC 2026 | |
| SU019 | Agibot (Genie Studio) | Genie Studio — End-to-End Embodied AI Development Platform | |
| SU020 | AGIBOT Store (official) | Agibot OmniHand O10 Product Listing | |
| SU021 | Humanoids Daily | The Great Numbers War: Unitree Claims Global Shipment Lead, Rebuts Agibot's No.1 Title | Unitree Robotics, which claimed 5,500 humanoid shipments … immediately contested Agibot's #1 ranking |
| SU022 | Pandaily | Agibot Completes New Funding Round, Valuation Exceeds 7 Billion Yuan | |
| SU023 | Humanoids Daily | China's Robot Fever Meets Reality: Beijing Warns of Bubbles and Repetitive Clones | Beijing warns of bubbles and repetitive clones in the humanoid robotics space |
| SU024 | Yicai Global | South Korea's LG Electronics, Mirae Asset Co-Invest in Chinese Humanoid Robot Startup Agibot | |
| SU025 | IDC | Humanoid Robotics: The Commercialization Tipping Point in 2026 | |
| SR001 | Bureau of Industry and Security (BIS), U.S. Department of Commerce | Homepage — Export Administration Regulations and Advanced Computing Guidance | BIS extended the Authorized IC Designer timeline until December 31, 2026; a license is required to export advanced computing items to entities headquartered in Country Group D:5 or Macau. |
| SR002 | Hill Dickinson LLP | Humanoid robots and the law — preparing for a new era of risk | Who is responsible if a robot causes harm — the manufacturer, the operator, or the software provider? Current product liability laws provide some guidance, but as robots evolve towards more autonomous decision making, new layers of complexity emerge. |
| SR003 | European Commission — Digital Strategy | A Risk-based Approach — EU AI Act regulatory framework for AI | AI safety components in critical infrastructures (e.g. transport), the failure of which could put the life and health of citizens at risk, are classified as high-risk; high-risk AI systems are subject to strict obligations before they can be put on the market. |
| SR004 | U.S. Occupational Safety and Health Administration (OSHA) | Robotics — Safety and Health Topics | There are currently no specific OSHA standards for the robotics industry. |
| SR005 | Office of Foreign Assets Control (OFAC), U.S. Department of the Treasury | OFAC Sanctions Programs and Information | |
| SR006 | International Organization for Standardization (ISO) | ISO 10218-1:2011 — Robots and robotic devices — Safety requirements for industrial robots (Withdrawn) | ISO 10218-1:2011 does not apply to non-industrial robots. Status: Withdrawal of International Standard [95.99]. |
| SR007 | National Institute of Standards and Technology (NIST) | Artificial Intelligence — NIST AI Risk Management and Standards | NIST promotes innovation and cultivates trust in the design, development, use and governance of AI technologies. NIST has a nonregulatory measurement science mission. |
| SR008 | International Organization for Standardization (ISO) | ISO/TS 15066:2016 — Robots and robotic devices — Collaborative robots | |
| SR009 | U.S. News & World Report (Reuters wire) | US Lawmakers to Introduce Bill to Ban Government Use of Chinese Robots | Two U.S. senators planned to introduce a bill that would ban the government from buying or operating humanoid robots made by Chinese firms; Agibot and Unitree are explicitly named. |
| SR010 | Humanoids Daily | China's Robot Fever Meets Reality: Beijing Warns of Bubbles and Repetitive Clones | Over 150 humanoid robot companies now operating in China; NDRC warns of 'speed and bubbles' and 'low-level redundancy' in humanoid sector. |
| SR011 | Humanoids Daily | China Drafts Dream Team for Humanoid Robot Standards: Taps Founders of Unitree and Agibot to Lead | 65-member MIIT committee led by Wang Xingxing (Unitree) and Peng Zhihui (Agibot) as deputy chairs. |
| SR012 | Humanoids Daily (citing Morgan Stanley AlphaWise survey) | Morgan Stanley Survey: Chinese Companies Want Humanoids But the Products Aren't Ready | 62% of Chinese C-suite respondents likely to adopt within 3 years; 23% satisfied with current products; 92% say unit cost must fall below 200,000 yuan. |
| SR013 | Humanoids Daily | The Great Numbers War: Unitree Claims Global Shipment Lead, Rebuts Agibot's No. 1 Title | Production figures in this industry are not audited and cannot be independently verified. |
| SR014 | Humanoids Daily | Unitree Files for $580M IPO — Humanoid Sales Surpass Robot Dogs as Profits Soar | Unitree IPO accepted by SSE; raises 4.202B RMB (~$580M); 2025 revenue >1.7B RMB ($235M), up 335% YoY; net profit >600M RMB ($83M), up 674%. |
| SR015 | IDC | Humanoid Robotics Commercialization 2026 | Accuracy of 80-90% in simulated environments often drops below 50% in real scenarios; sim-to-real gap remains a bottleneck. |
| SR016 | IEEE Spectrum | The Humanoid Robot Revolution Is Coming — Eventually | Collaborative robots took seven to ten years from first commercial versions to batch sales; no specific OSHA standards for robotics; ISO 10218 designed for fixed industrial robots. |
| SR017 | Robotics and Automation News | Agibot Secures Strategic Investment from LG Electronics and Mirae Asset | Agibot investors include BYD, SAIC, BAIC, JD.com — state-owned enterprise exposure creates governance and CFIUS risk. |
| SR018 | International Federation of Robotics (IFR) | China Makes AI-Powered Robots Core of National Strategy | Mass adoption as universal humanoid factory helpers in private households will not happen within near- and medium-term future; traditional industrial robots faster and more reliable for repetitive precision tasks. |
| SR019 | Interesting Engineering | Agibot G2 Humanoid Robots Go Live on Production Line at Longcheer | G2 robot 99.9%+ success rate at Longcheer — single deployment site under controlled conditions; scalability to multi-facility not verified. |
| SR020 | Yicai Global | South Korea's LG Electronics, Mirae Asset Co-invest in Chinese Humanoid Robot Startup Agibot | AgiBot has completed 10 financing rounds, with investment from Tencent Holdings, JD.Com, BYD, SAIC Motor, BAIC Group, and TCL. |
| SR021 | Humanoids Daily | AGILINK Hits Unicorn Status: AGIBOT's Dexterous Hand Spin-off Raises Hundreds of Millions | |
| SR022 | Taibo Financial (translating Goldman Sachs China survey) | Goldman Sachs Survey of 8 Chinese Humanoid Robot Makers: 2025-2026 Outlook | Simulate to Reality gap still a bottleneck; accuracy drops from 80-90% in simulation to below 50% in real scenarios; time required for large-scale real data collection limits deployment speed. |
| SR023 | Pandaily | Agilink Launches OmniHand 3 Series, Ships Over 8,000 Dexterous Hands | |
| SR024 | Robotics and Automation News | Why China's New Humanoid Robot Standards Could Change the Industry | China's MIIT published its first national standard system for humanoid robots and embodied intelligence in late February 2026. |
| SR025 | Humanoids Daily | Deployment Year One: Agibot Unveils Massive Fleet and AI Model Stack at APC 2026 | |
| SR026 | Goldman Sachs | Humanoid Robots — Goldman Sachs Research Overview | Humanoid robot solutions could be economically viable in factory settings between 2025 to 2028; robots must bring down production costs ~15-20% a year. |
| SR027 | Humanoids Daily | UBTECH Claims World's First Mass Delivery of Humanoids as Robot Army Deploys | UBTECH 2025 order book exceeded 800M yuan ($113M); automotive and logistics identified as core buyer segments. |
| SR028 | Crunchbase News | New Unicorn Startups — May 2026: OpenAI, Anthropic, SpaceX, Robotics | Shanghai-based Agilink has raised four funding rounds since it was spun out of Agibot in January, and reached a valuation of $1 billion. |
| SR029 | Reuters | China's Agibot Spinoff Agilink Valued at $1 Billion in Latest Funding Round | |
| SR030 | MarketsandMarkets | Humanoid Robot Market — Global Forecast to 2030 | |
| SV001 | Crunchbase News | New Unicorn Startups of May 2026: Agilink and others reach $1B valuations | Shanghai-based Agilink has raised four funding rounds since it was spun out of Agibot in January, and reached a valuation of $1 billion. Agilink is focused on dexterous hand technology. |
| SV002 | TechCrunch | Figure reaches $39B valuation in latest funding round | Figure raised a Series C funding round that values it at $39 billion, exceeding $1 billion in total capital raised since 2022. |
| SV003 | PR Newswire (Figure AI) | Figure Raises $675M at $2.6B Valuation and Signs Collaboration Agreement with OpenAI | Figure has raised $675M in Series B funding at a $2.6B valuation with investments from Microsoft, OpenAI Startup Fund, NVIDIA, Jeff Bezos, ARK Invest. |
| SV004 | Apptronik | Apptronik Closes Over $935 Million Series A | $520M Series A-X bringing Apptronik's total Series A to more than $935M and total capital raised to nearly $1B. |
| SV005 | TechStartups | Apptronik raises $520M at $5B valuation to bring humanoid robots to market ahead of Tesla | $5B valuation; Howard Morgan expects demand for Apollo to reach $1B in orders starting in 2027, with robots priced at roughly $80,000 per year. |
| SV006 | Crunchbase News | Robotics Startups Raised Nearly $14B in 2025; China Tops $5.6B by Mid-May 2026 | Startups in the sector raised nearly $14 billion in funding in 2025, up from $8.2 billion in 2024. China-based robotics companies raised $5.6 billion in venture investment through mid-May 2026. |
| SV007 | Humanoids Daily | Unitree Files for $580M IPO — Humanoid Sales Surpass Robot Dogs as Profits Soar | 2025 revenue >1.7B RMB ($235M), up 335% YoY; net profit >600M RMB ($83M), up 674%; 5,500 humanoid units shipped; $580M IPO raise on SSE STAR Market. |
| SV008 | Pandaily | AgiBot Completes New Funding Round, Valuation Exceeds 7 Billion Yuan | Valuation Exceeds 7 Billion Yuan (~$960M USD at August 2025 exchange rates). |
| SV009 | Yicai Global | South Korea's LG Electronics and Mirae Asset Co-invest in Chinese Humanoid Robot Startup AgiBot | AgiBot has completed 10 financing rounds, with investment from big firms such as Tencent Holdings, JD.Com, BYD, SAIC Motor, BAIC Group, and TCL. |
| SV010 | Sacra | Figure AI valuation, funding and news | Figure AI reached a $39 billion post-money valuation in September 2025 following a Series C funding round that exceeded $1 billion. |
| SV011 | Humanoid Index | Figure AI: Funding, Valuation, Robot Specs and More | |
| SV012 | Humanoid Index | Agility Robotics: Funding, Valuation, Robot Specs and More | |
| SV013 | Humanoid Index | Apptronik: Funding, Valuation, Robot Specs and More | |
| SV014 | 1X Technologies | 1X Secures $100M in Series B Funding | $100M Series B; NEO Beta units to be sent to select homes in late 2024; OpenAI Startup Fund as lead. |
| SV015 | TechCrunch | OpenAI-backed 1X raises another $100M for the race to humanoid robots | |
| SV016 | US Securities and Exchange Commission | Form D — HII Apptronik Series I, a Series of HII Apptronik LLC | HII Apptronik Series I, a Series of HII Apptronik, LLC — filed 2025-11-14 with the SEC as a Form D exempt offering, confirming institutional investment vehicle for Apptronik Series A extension. |
| SV017 | Goldman Sachs Research | Global Automation: Humanoid Robot — The AI Accelerant | |
| SV018 | IDC | Humanoid Robotics Commercialization in 2026 | Global shipments of humanoid robots are expected to exceed 510,000 units by 2030 at CAGR of ~95%. Chinese vendors account for 95% of 2025 shipments. |
| SV019 | International Federation of Robotics | China Makes AI-Powered Robots Core of National Strategy | China's manufacturing industry already has an operational stock of around 2 million units — ~4.5x Japan. 54% of annual industrial robots installed worldwide deployed in China. |
| SV020 | Humanoids Daily | The Great Numbers War: Unitree Claims Global Shipment Lead, Rebuts AgiBot's No. 1 Title | Production figures in this industry are not audited and cannot be independently verified. Unitree vs. Agibot dispute over shipment counts (5,500 vs. 5,168) highlights IPO-race incentives to inflate claims. |
| SV021 | Humanoids Daily | China's Robot Fever Meets Reality: Beijing Warns of Bubbles and Repetitive Clones | Over 150 humanoid robot companies now operating in China. NDRC warns of 'speed and bubbles' and 'low-level redundancy' in humanoid sector. |
| SV022 | Humanoids Daily | Morgan Stanley Survey: Chinese Companies Want Humanoids But the Products Aren't Ready | 62% of Chinese C-suite respondents likely to adopt humanoids within 3 years; 23% satisfied with current products; 92% say unit cost must fall below 200,000 yuan ($28K). |
| SV023 | Taibo Financial | Goldman Sachs January 2026 Survey: 8 Chinese Humanoid Robot Makers — Verification Window | Goldman Sachs expects global shipments to reach approximately 15,000-20,000 units by 2025, with Chinese companies contributing the majority; 2026 as 'volume verification + expectation reset'. |
| SV024 | Forbes | Humanoid Robots: Here Are The 16 Leading Manufacturers | Goldman Sachs says selling humanoid robots will be a $38 billion space by 2035, while ARK Invest says the resulting economic value of their labor could be as high as $24 trillion. |
| SV025 | Robotics and Automation News | AgiBot Secures Strategic Investment from LG Electronics and Mirae Asset | |
| SV026 | Humanoid Index | AgiBot: Funding, Valuation, and More | Omdia ranks AgiBot #1 in humanoid shipments 2025 (>5,100 units); total disclosed funding $83.8M per Tracxn. |
| SV027 | Shanghai Stock Exchange (SSE) | Unitree Robotics Co., Ltd. — Initial Public Offering Application Document Disclosure (002178) | SSE STAR Market IPO application filed March 20, 2026 — primary disclosure document for Unitree Robotics; includes audited 2025 financials and IPO terms. |
| SV028 | IEEE Spectrum | Humanoid Robots — The Long Road to Commercialization | |
| SV029 | Humanoids Daily | China Drafts Dream Team for Humanoid Robot Standards — Taps Founders of Unitree and AgiBot to Lead | |
| SV030 | Goldman Sachs | Humanoid Robots — Goldman Sachs Research Summary | Humanoid robot solutions could be economically viable in factory settings between 2025 to 2028; robots must bring down production costs ~15-20% a year. |