Rhoda AI
技术底子强、融资充足的物理 AI 竞争者,但商业披露很薄。
physical-AI 逻辑有吸引力、背书顶级,但商业披露太少,难以有把握地支撑 $1.7B 标记估值。
封面要素
公司概况
Rhoda AI 是一家位于 Palo Alto 的机器人智能创业公司,正在打造 FutureVision:一个硬件无关的智能层,使用 Direct Video Action 世界模型来自动化变化较大的工业工作流。公开资料看,公司于 2024 年完成法律设立,经过 18 个月隐身期后在 2026 年 3 月带着 $450M A 轮亮相,目标是落地制造和物流部署,但收入、客户名称和董事会结构仍保持私密。
- 成立时间
- 2024-08-01
- 创始人
- Jagdeep Singh, Gordon Wetzstein
- 创立地点
- Palo Alto, CA
- 总部
- Palo Alto, CA, USA
- 产品
- FutureVision 是建立在 Direct Video Action 模型上的机器人智能层,先用互联网规模视频做预训练,再用相对少量机器人数据适配工业任务。
- 客户
- 汽车、制造、物流和电商工作流中的工业企业,需要自动化变化较大的物料搬运任务。
- 商业模式
- 企业机器人智能授权与部署支持;FutureVision 计划运行在 Rhoda 系统,以及合作方硬件 / 软件平台之上。
- 阶段
- Series A
- 融资情况
- 2026 年 3 月宣布 $450M A 轮,估值据报道约 $1.7B。
执行摘要
主要优势
- 独特的 DVA/FutureVision 架构,以互联网规模视频预训练和闭环控制为锚。
- 对早期 physical-AI 公司来说,投资方组合异常强,Series A 规模也很大。
- 工业工作流聚焦可信,覆盖汽车、制造、物流和电商用例。
主要风险
- 估值已经很大,但没有公开收入、定价、毛利率或具名客户数据。
- 真实世界稳健性、安全性和部署可重复性,在公开证据中只得到部分验证。
- 如果同业无法把试点转成披露的软件经济性,physical-AI 私募市场估值可能压缩。
未决问题
- 具名付费客户、部署数量、续约情况和客户访谈。
- 定价模型、软件抽成率、毛利率,以及任何 ARR 或收入基数。
- 准确的董事会 / 治理结构,以及法定注册日期是否匹配完整运营创立时间线。
目录
01公司概况
1.1 身份、产品与运营模式
Rhoda AI 经过 18 个月隐身期后,于 2026 年 3 月 10 日公开亮相,给自己的定位是位于 Palo Alto 的机器人智能公司,目标是把通用型机器人从实验室演示推进生产环境。官网、新闻材料和发布报道里的口径一致:FutureVision 是商业化智能层,Direct Video Action(DVA)是底层技术押注。卖点不只是“更好的机器人控制策略”,而是一套不同的学习栈:Rhoda 先用互联网规模视频做预训练,再用较少机器人数据做后训练,让系统能在传统工业自动化难以处理的高变化环境中泛化。 运营模式看起来也更像混合型,而不是纯软件或纯硬件。Rhoda 称 FutureVision 计划随时间授权给合作方硬件和软件平台,但首页同时在推广 Rhoda 机器人平台,带有定制执行器、安全等级视觉和载荷能力说法,暗示内部也在做硬件。公开示例集中在工业操作而非消费机器人:退货处理、汽车产线轴承拆袋转移、重型容器拆解,以及跟随演示执行任务。这给公司提供了可信的制造 / 物流切入点,但也意味着业务必须证明,技术演示能转成可重复的商业部署,而不是一次性试点胜利。[CO001, CO002, CO003, CO004, CO005, CO006]
| 指标 | 数值 / 状态 | 日期 | 置信度 | 缺口 / 备注 |
|---|---|---|---|---|
| 运营总部表述 | Palo Alto, CA | 2026-03-10 | 高 | 官方发布材料使用 Palo Alto;注册数据则显示 San Jose 注册地址 |
| 法律实体备案 | Rhoda AI Corporation 于 2024-08-01 在 Delaware 注册;在 California 处于存续状态 | 2024-08-01 | 中 | 注册信息来源次于政府检索,但提供备案历史 |
| 当前阶段 | Series A 轮 | 2026-03-10 | 高 | 部分追踪平台误分为 Series B;官方来源称为 Series A |
| 发布融资 | 已宣布 $450M | 2026-03-10 | 高 | 得到官方新闻稿、Business Wire 和法律报道交叉印证 |
| 估值 | 据报 ~$1.7B | 2026-03-10 至 2026-03-11 | 中 | 估值出现在二级报道中,而非 Rhoda 自己的新闻稿 |
| 公开团队名单 | 团队页列出 62 名具名人员 | 2026-06-09 | 中 | 只是下限,不是完整员工普查 |
| 开放职位 | 公开 Ashby 招聘板有 33 个岗位 | 2026-06-09 | 中 | 招聘覆盖面显示在快速扩编,但不等于净员工数 |
| 收入 / ARR | null | 2026-06-09 | 低 | 检索到的来源未披露公开收入或 ARR |
| 具名客户 / 客户数 | null | 2026-06-09 | 低 | 公司提到工业合作伙伴,但未具名客户 |
| 公开定价 | null | 2026-06-09 | 低 | 未披露公开定价、ACV 或合同结构 |
混合列示已确认的公开事实和明确披露缺口;null 表示所审阅来源未公开披露该指标。
[CO002, CO020, CO021, CO022, CO024, CO025]这张图展示 Rhoda 如何把互联网级视频学习、FutureVision、工业工作流、硬件选择和资本串成一个业务系统。
[CO003, CO004, CO005, CO008, CO011, CO012]截至本报告生成日,可由公开信息支撑的成熟度信号;不含收入、客户数等未披露指标。
只使用已明确披露或可从已检索公开记录直接观察的指标;有意排除未披露的收入和客户数。
[CO010, CO012, CO019, CO020, CO024, CO025]1.2 领导梯队与关键人物集中度
公开可见的领导层集中在少数人身上。Jagdeep Singh 作为 CEO 和联合创始人,是公司的门面;技术叙事由首席科学家 Eric Chan 和 Stanford 教授 Gordon Wetzstein 承载;更宽的公开领导团队包括产品、研究、战略和数据高管。这足以说明 Rhoda 不只是两个人的科研项目,但也暴露了公开控制面偏窄:投资人被要求承销的故事,仍高度绑定 Singh 的创始人口碑,以及 Chan 和 Wetzstein 的研究可信度。 现有资料也显示,研究实力和公开可见部署深度之间并不对称。Wetzstein 的 Stanford 页面确认他自 2024 年 10 月起就是 Rhoda 联合创始人,把公司锚定在严肃的学术视频生成能力上。Chan 被描述为 Stanford 研究员、前 WorldLabs 生成模型架构师。但检索到的材料没有披露董事会,没有讲清治理结构,也没有呈现具备大规模部署资历的现场服务或工业运营负责人。这个缺口很关键,因为 Rhoda 瞄准的是变化多的制造和物流工作流,安装、安全、正常运行时间和售后执行可能与模型质量同样重要。招聘板和团队页显示公司确实在广泛招人,但仍没有消除公开公司画像里的关键人物依赖。[CO013, CO014, CO015, CO016, CO017, CO018]
| 人物 | 职务 | 公开背景 | 创始人与市场匹配 / 覆盖 | 关键人物依赖 |
|---|---|---|---|---|
| Jagdeep Singh | CEO、联合创始人 | 连续深科技创业者;发布时的公开代表 | 商业叙事者,也可能是资本配置者 | 高 — 创始人口碑锚定公司叙事 |
| Eric Ryan Chan | 首席科学家 | Stanford 研究员;前 WorldLabs 生成模型架构师 | 连接前沿视频生成与机器人学习栈 | 高 — 技术可信度的核心 |
| Gordon Wetzstein | 联合创始人 / 科学顾问 | Stanford EE 教授;Stanford Physical and Spatial Intelligence Lab | 学术可信度与视频 / 世界模型专长 | 高 — 公开技术信任部分建立在其履历上 |
| Andrew Wooten | 首席产品官 | 团队页具名 | 产品和商业化接口 | 中 — 角色重要,但公开信息较少 |
| Changan Chen | 首席研究官 | 团队页具名 | Chan/Wetzstein 之外的研究执行宽度 | 中 — 扩大团队纵深,但公开背景稀疏 |
| Steve Tirado | 首席战略官 | 团队页具名 | 战略 / 外部定位支持 | 中 — 公开职责可见,详细履历未披露 |
| Alex Bergman | 首席数据官 / 软件工程 VP | 团队页具名 | 数据和软件执行覆盖 | 中 — 重要内部建设者角色,公开细节有限 |
只完整覆盖 Rhoda 已检索页面中可见的公开具名领导岗位;不代表完整管理组织。
[CO013, CO014, CO015, CO016, CO017, CO018]1.3 融资基础、投资方与法律足迹
Rhoda 公开亮相时同步宣布了一笔融资;对于刚走出隐身期的公司来说,规模非常大。官方发布材料和法律顾问报道相互印证:公司于 2026 年 3 月 10 日宣布 $450M A 轮;多家二级来源将估值放在约 $1.7B。投资方名单广且质量高,包括 Premji Invest、Khosla Ventures、Temasek、Mayfield、Capricorn、Prelude Ventures、Xora、John Doerr 等。这个投资团既给资本,也给战略信号,但并不能完全说明本轮领投权究竟有多集中。 领投归属正好说明,后续承销应依赖签署版融资文件,而不是媒体摘要。Rhoda 自己的新闻稿列出支持方,但没有指定单一领投。Wilson Sonsini 称本轮由多名机构组成的投资团领投。几篇二级文章则把 Premji Invest 称为领投方,部分第三方追踪平台甚至误把本轮标成 B 轮。法律层面,California 注册数据显示 Rhoda AI Corporation 是一家有效存续的 Delaware 公司,成立于 2024 年 8 月 1 日,注册地址在 San Jose。这不一定与 Palo Alto 运营总部矛盾,但说明公开法律足迹和公开运营地点口径并不完全一致。重要结论是:公司资本基础很强,但围绕阶段、领投方和设立细节的公开元数据,比标题报道暗示的更嘈杂。[CO021, CO022, CO024, CO025, CO026, CO027]
| 利益相关方 | 角色 | 控制权 / 经济重要性 | 尽调问题 |
|---|---|---|---|
| Premji Invest | 具名支持方;二级报道常称其为领投方 | 潜在领投或锚定投资者,但官方材料未确认其单独领投 | 获取已签署投资条款书 / 股权结构表,确认领投身份和持股 |
| Khosla Ventures | 具名支持方 | 重要机器人 / AI 战略信号 | 确认出资规模和任何治理权利 |
| Temasek | 具名支持方 | 国际主权资本信号及可能的亚洲工业网络 | 核查战略商业预期和董事会观察员权利 |
| Mayfield | 具名支持方 | 老牌风投支持方;机器人经济学评论中被引用 | 确认 Mayfield 是否拥有董事席位或信息权 |
| Capricorn Investment Group | 具名支持方 | 出现在官方支持方名单和 WSGR 轮次说明中 | 澄清经济权益以及 Capricorn 是否共同领投 |
| Prelude Ventures | 具名支持方,且在其投资组合中列示 | 气候 / 前沿科技主题投资者;投资组合页面确认持有关系 | 询问运营支持是否延伸到工业伙伴 |
| John Doerr | 个人投资者 | 高信号个人支持方,带来网络价值 | 澄清经济持股与信号价值的差别 |
| Leitmotif / Matter / Xora | 具名支持方 | 更广泛财团的一部分;可能补充汽车 / 工业网络 | 确认参与规模和战略角色 |
仅为公开具名投资者和利益相关方的部分地图;持股比例、董事会权利和轮次经济性仍未披露。
[CO024, CO025, CO026, CO027, CO028, CO029]梳理从成立证据到发布、融资、首次公开可见的质疑,再到当前招聘强度的公开时间线。
2024 年备案前的成立时间未被公开证据锚定;因此时间线从第一条可靠公开证据开始。
[CO001, CO010, CO016, CO019, CO021, CO024]1.4 里程碑、证据点与未披露事项
Rhoda 的里程碑时间线很压缩。公开记录从 2024 年法律设立证据,直接跳到 2026 年发布:一场头部融资事件、一份旗舰新闻稿和密集招聘。官网和转载报道提供了足够证据,说明公司关注的是真实工业任务,而不是投机式机器人品牌包装:它展示了具体物流和汽车工作流,声称在一次高产量评估中制造周期低于两分钟,并挂出 33 个开放职位,集中在研究、软件和硬件。这些都是有意义的就绪信号,尤其对于一家刚从隐身期出来的创业公司。 但公开披露包在尽调最需要硬证据的地方仍然偏薄。公司没有点名客户,不公布收入或客户数,不披露定价,也没有给出确切员工数,只有团队页姓名、招聘职位和外部数据平台估算等间接代理指标。新闻页目前指向单一的 2026 年 3 月新闻稿,进一步说明沟通记录仍很窄。robotics.press 的第三方分析更进一步,认为投资人在没有独立验证部署或经济性的情况下承销强技术逻辑。这个批评可能放大了下行,但准确抓住了主要尽调约束:Rhoda 目前最强的公开信号是技术架构、投资方质量和招聘动能,而不是已验证商业牵引力。[CO001, CO010, CO019, CO023, CO033, CO034]
| 日期 | 事件 | 类型 | 金额 / 估值 / 状态 | 参与方 | 含义 |
|---|---|---|---|---|---|
| 2024-08-01 | 法律实体成立,并登记于 California 记录 | 创立 | Delaware 公司;在 CA 处于存续状态 | Rhoda AI Corporation | 最早的具体公开成立证据 |
| 2024-10-01 | Wetzstein 公开资料显示其自 Oct 2024 起为 Rhoda 联合创始人 | 治理 | 科学联合创始人角色可见 | Gordon Wetzstein | 锚定创始科学家时间线 |
| 2026-03-10 | 公司走出隐身模式并公开发布 | 产品 | FutureVision 发布 | Rhoda AI | 形成第一条公开经营记录 |
| 2026-03-10 | 宣布 Series A 融资 | 融资 | 已宣布 $450M | 投资财团包括 Premji、Khosla、Temasek、Mayfield | 给 Rhoda 带来异常充足的初始弹药 |
| 2026-03-10 | DVA 架构公开披露 | 产品 | 视频优先的闭环控制 | Rhoda 研究 / 发布团队 | 技术差异化变得明确 |
| 2026-03-10 | 披露制造业基准测试 | 规模 | 大批量评估中周期 <2 分钟 | Rhoda + 未具名工业合作方 | 最好的公开商业化证据点 |
| 2026-03-10 | 官方新闻页发布第一篇也是唯一可见文章 | 治理 | 网站上只有一篇发布新闻稿 | Rhoda 传播团队 | 显示公开传播历史很窄 |
| 2026-03-11 | 二级报道称估值 ~$1.7B | 融资 | 据报估值 ~$1.7B | 多家媒体 | 表明在公开收入披露前,投资者已经愿意支付高价 |
| 2026-03-13 | robotics.press 发布悲观情景分析 | 反向 | 强调执行和披露风险 | 独立分析网站 | 在公开记录中引入第一条可见质疑 |
| 2026-06-09 | 招聘板显示 Palo Alto 有 33 个开放职位 | 规模 | 招聘覆盖研究、软件、硬件、运营 | Rhoda 招聘团队 | 显示发布后仍在积极扩建 |
仅为检索到的公开来源形成的部分时间线;确切成立日期、董事会里程碑和客户里程碑仍未披露。
[CO001, CO010, CO016, CO019, CO021, CO023]02市场分析
2.1 市场边界:Rhoda 卖的是智能层,不是整套机器人栈
Rhoda 自己的发布材料一直把 FutureVision 描述为机器人智能系统,而不是机器人 OEM 产品。公司称 Direct Video Action 架构是一个“智能层”,今天可驱动 Rhoda 系统,之后可授权到不同机器人硬件和软件平台。这一点重要,因为它大幅收窄相关市场边界。正确比较集合不是整个工业机器人硬件、传感器或工厂自动化资本支出,而是软件和模型层:硬件已经存在或正在采购后,正是这一层让高变化、高混合度的实体工作流变得可自动化。 这个区分很关键,因为多数分析师的机器人市场仍把硬件重的类别打包在一起。MarketsandMarkets 的 AI 机器人市场预测,2025 年硬件仍占支出 61%,同时软件和服务扩张。同样,其物理 AI 市场分类把 GPU、传感器、内存、执行器等硬件组件和软件、服务并列。对 Rhoda 来说,这些硬件重数字可作为自上而下背景,也能证明装机基础足够大,足以支撑一个软件层,但它们不是 Rhoda 能直接捕获的收入。 公开描述也指向 Rhoda 的第一批滩头阵地:制造和物流环境,材料、布局和工作流持续变化。在这些场景里,既有替代方案不只是人工,还包括固定自动化、定制机器人编程、重度依赖远程操作的训练闭环,以及一旦工作流变化就变贵的系统集成项目。Rhoda 的切入点,是主张闭环、视频训练模型能降低机器人适配新任务的边际成本。因此,相关纳入支出包括:机器人智能软件、世界模型或策略模型授权、任务适配、编排和相关集成。排除支出则包括机器人臂、夹爪、移动底盘、货架、输送线,以及占据更广泛仓库和工厂自动化预算的大部分场地建设支出。[CM001, CM002, CM003, CM004, CM005, CM006]
| 细分 / 类别 | 纳入支出 | 排除支出 | 买方 / 付款方 | 对 Rhoda 的意义 |
|---|---|---|---|---|
| 机器人智能层 | 基础模型授权、策略更新、世界模型推理、适配工具 | 机械臂、夹爪、底盘、存储系统 | 运营 / 自动化预算负责人 | 最贴近 Rhoda 声称的授权模式 |
| 部署软件与编排 | WES / 编排、任务排序、异常处理、分析 | 货架建设、输送线、托盘硬件 | 供应链工程 / 仓库运营 | 存量场地中最可能的挂载点 |
| 数据 / 训练栈 | 减少遥操作、仿真、数据管线、模型改进 | 工厂建设、网络升级、通用云 | 创新 / 高级自动化团队 | 帮助证明单次试点之外的软件抽成合理性 |
| 系统集成邻近项 | 实施、支持、工作流重设计、托管服务包装 | 总包工程和非机器人场地工程 | 集成商加企业发起人 | 重要进入市场路径,但不会都是 Rhoda 收入 |
| 现状替代方案 | 人工作业、固定编程、按任务定制 ML、传统自动化软件 | N/A | 现有运营负责人 | Rhoda 必须替代或补充的预算 |
纳入支出聚焦软件和智能层捕获。更广的硬件和场地资本开支类别仅作为预算归属和采用路径的背景。
[CM002, CM003, CM004, CM005, CM006, CM030]Rhoda 最能站住脚的市场堆栈,从宽泛自动化预算收窄到更小的智能层切口。
金字塔是概念图,不可相加。它反映范围逐层收窄,而不是经审计的市场份额计算,因为 Rhoda 未披露定价或抽成率。
[CM003, CM010, CM014, CM015, CM022, CM048]2.2 规模测算视角:增长真实存在,但直接 SAM 小于机器人 TAM 标题数字
多个规模测算视角都支持 Rhoda 的建设性背景,但每个视角回答的问题不同。最宽的软件加硬件视角是 AI 机器人市场,MarketsandMarkets 预计 2025 年为 $6.11B,2030 年达到 $33.39B。更窄的“物理 AI”视角落在 2026 年 $1.50B,并到 2032 年增长至 $15.24B。再往外,Mordor Intelligence 估算 2026 年仓库自动化市场为 $34.17B,Modern Materials Handling 指出 2023 年全球仓库自动化投资约 $21B,到 2033 年超过 $90B。这些数字不能相加,但合在一起呈现一层栈:顶部是巨大的物流和工厂运营,下面是大型自动化预算,内部则是更小但快速增长的智能层。 装机基础视角同样重要。IFR 称 2024 年安装了 542,000 台工业机器人,全球已有 4.664 million 台在运行;预计 2025 年为 575,000 台,2028 年超过 700,000 台。这意味着市场不再单纯受限于“机器人会不会存在”。商业问题转向:通用化智能软件能在哪些地方提高已部署机队的价值,或让新类别工作变得可自动化。 对 Rhoda 而言,这意味着一套实际的 TAM/SAM/SOM 层级。TAM 是物理 AI 和可适配机器人控制支出中偏软件的切片。SAM 是制造、仓库和物流工作流里的子集,运营方既有变化性问题,也有预算权购买智能软件。SOM 更窄:愿意在定价、正常运行时间和安全主张完全验证前,先试点新模型驱动工作流的存量工业场地和系统集成商。由于公开来源没有披露 Rhoda 定价、转化率或抽成率,今天任何 SOM 都必须受证据约束并明确加注,而不能被当作精确预测。[CM010, CM011, CM014, CM015, CM016, CM017]
| 视角 | 发布方 / 年份 | 地区 | 数值 | 增长 | 方法 / 范围 | 置信度 | 局限 |
|---|---|---|---|---|---|---|---|
| AI 机器人市场(广义) | MarketsandMarkets / 2025 | 全球 | $6.11B (2025) → $33.39B (2030) | 40.4% CAGR | 软件 + 硬件 AI 驱动机器人栈 | 中 | 仍偏硬件;并非 Rhoda 直接收入 |
| 具身 AI 市场(较窄) | MarketsandMarkets / 2026 | 全球 | $1.50B (2026) → $15.24B (2032) | 47.2% CAGR | 包括软件、服务和机器人相关部件的具身 AI 产品 | 中 | 除软件外还包括半导体 / 传感器 / 执行器 |
| 仓库自动化市场 | Mordor Intelligence / 2026 | 全球 | $34.17B (2026) → $65.74B (2031) | 13.98% CAGR | 完整仓库自动化系统 | 中 | 主要是基础设施和硬件,加上软件 |
| 仓库自动化投资 | Modern Materials Handling / 2026 | 全球调研 / 基准 | $21B (2023) → >$90B (2033) | 10 年增长 329% | 仓库自动化已观察和预测支出 | 中 | 投资趋势,不是纯软件 TAM |
| 3PL 需求环境 | StartUs Insights / 2026 | 全球 | $1.8T (2026) → $4.3T (2035) | 10.1% CAGR | 推动自动化需求的底层物流工作流池 | 中 | 运营市场规模,不等同于自动化可获取份额 |
| 装机基数视角 | IFR / 2025 | 全球 | 在役机器人 4.664M 台;2024 年安装 542k 台 | 2025 年预计安装 575k 台 | 以台数衡量可触达机器人基数 | 中 | 台数要转成美元,还需要软件渗透率假设 |
这些口径本来就不该相加。它们从物流运营一路收窄到 Rhoda 瞄准的智能层,展示不同宽窄的市场范围。
[CM010, CM011, CM014, CM015, CM016, CM017]从宽泛自动化到更窄的物理 AI 层,用区间展示与 Rhoda 有关的主要市场视角。
中值是插值视觉锚点,并非出版方估计。该图比较不同视角的规模差异,不应相加。
[CM010, CM014, CM015, CM016, CM048]Rhoda 的采用机会从巨大的运营市场,收窄到近期小得多的软件切口。
所有层级单位均为 USD 十亿美元。最后的切口是分析师估计,只用于展示 Rhoda 直接可获取市场应比包含硬件的 TAM 小多少。
[CM014, CM015, CM036, CM048, CM049, CM050]2.3 买方、用户与付款方:采用路径要穿过既有运营预算和集成商渠道
Rhoda 的买方地图由谁已经掌握自动化预算来决定。在制造单元里,日常用户可能是自动化工程师、产线主管或机器人集成商,经济买方则在工厂运营、先进制造或中央自动化负责人手里。在仓库里,用户更常是机器人或履约运营团队,但付款方在网络运营、供应链工程或物流领导层手里,他们已经管理 WMS、WES、AMR 和系统集成支出。这个分工很重要,因为 Rhoda 的“机器人脑”价值主张,只有在既有资本支出或运营支出预算框架内提升吞吐、正常运行时间、劳动生产率或部署灵活性时,才会拿到预算。 公开买方优先级数据说明这些团队在乎什么。Modern Materials Handling 称,耐用性、可靠性和正常运行时间主导选型标准,其后是快速服务响应、采购价、总拥有成本,以及与现有设备集成。McKinsey 的工业调查补充说,许多客户偏好全服务实施模式,同时担心资本成本和内部经验不足。Hy-Tek 的 2026 年仓库趋势文章进一步强化:软件编排,尤其是 WES 和 low-code 集成,已经成为核心,因为企业越来越需要异构系统像一套栈一样工作。 这为 Rhoda 勾勒出一条可能采用路径。第一笔销售很少会是一张白纸上卖出的纯基础模型授权。更常见的是,作为试点或存量场地增强,被插入既有自动化项目,往往经由 OEM、集成商或运营发起人。运营用户想要更少异常、更快任务适配。预算所有者想要可信 ROI。付款方想确认供应商能支撑部署,而不只是发布好看的 demo。因此,Rhoda 的商业化挑战不只是技术泛化;它还要在已经会因正常运行时间或安全不确定而惩罚新东西的采购流程里,赢得一席之地。[CM004, CM008, CM021, CM022, CM023, CM024]
| 细分市场 | 买方 | 用户 | 付费方 | 工作流 | 预算归属 | 采用触发点 |
|---|---|---|---|---|---|---|
| 高变化制造单元 | 工厂自动化负责人 | 自动化工程师 / 产线主管 | 工厂运营 | 拣选、配套、部件搬运 | 先进制造资本开支 | 劳动力短缺 + 换线复杂 |
| 仓库入库 / 拆垛 | 配送运营负责人 | 机器人 / 维护团队 | 仓配网络运营 | 入库托盘拆解、分拣、检查 | 仓库自动化预算 | 异常处理 + 吞吐瓶颈 |
| 仓库件拣 / 混合 SKU 处理 | 履约工程 | 机器人运营团队 | 供应链副总裁 / CFO | 混箱和单件操作 | WES / 自动化项目 | 人工痛点 + 服务水平压力 |
| 3PL / 系统集成商渠道 | 集成商总经理 | 解决方案架构师 | 集成商项目预算 | 将 Rhoda 式智能打包进客户项目 | 项目服务 + 软件转售 | 需要灵活差异化 |
| 工业 OEM / 伙伴授权 | OEM 产品负责人 | 嵌入式机器人软件团队 | OEM 研发 / 平台预算 | 给既有硬件加通用策略层 | 平台 / 产品预算 | 新任务上市更快 |
买方和付费方通常高于日常用户。Rhoda 必须先打进既有自动化项目,而不能默认一开始就有独立软件采购流程。
[CM004, CM008, CM022, CM025, CM030, CM031]不同工作流下,买方、用户和付款方关系不同,但所有路径都要经过现有自动化负责人。
单元格综合公开工作流描述和常见采购模式。Rhoda 未发布正式买方地图。
[CM004, CM022, CM025, CM030, CM031, CM032]2.4 增长驱动与约束:真实需求顺风存在,炒作和证据缺口也真实存在
Rhoda 这类软件的需求逻辑很直接。制造商和仓库仍依赖数百万人工岗位,BLS 预计物料搬运每年有超过 1 million 个职位空缺,U.S. Chamber 仍显示制造业有数十万开放岗位。NVIDIA 的 2026 年零售和 CPG 调查显示 AI 预算在扩张,UPS 和 DHL 都把 2026 年物流投资围绕韧性、可见性、软件定义仓库、AMR 和 AI 辅助决策展开。Interact Analysis 和 Modern Materials Handling 也都显示,即使宏观环境波动,自动化预算仍在前进。 同一批来源也说明,投资人为什么要抵抗物理 AI 炒作。McKinsey 仍把资本成本和内部经验不足视为主要阻碍。Interact Analysis 对移动机器人前景的下调幅度大于固定自动化,并称不确定性、关税和更高钢材成本正在扭曲项目时点。存量场地改造占主导,因为企业对新建项目承诺很谨慎。即便支持性来源,也越来越把市场框定在软件定义编排、服务质量和 ROI 纪律,而不是任何自称“物理 AI”的公司都能拿到不受约束的支出。 这带来两个明确限定。第一,许多公开市场数字仍偏硬件,而 Rhoda 能变现的层只是这些预算的一小片。第二,Rhoda 自身公开证据仍薄:没有披露收入、定价、具名付费客户名单,也没有按工作流公布抽成率。这不会否定市场机会,但意味着市场逻辑应被视为可信需求背景,而不是 Rhoda 已经抓住持久软件切入点的证明。换句话说,品类真实存在,但具体商业胜利仍是承销问题。[CM026, CM027, CM028, CM029, CM037, CM038]
| 驱动因素 / 约束 | 方向 | 时间 | 含义 | 尽调问题 |
|---|---|---|---|---|
| 制造业用工短缺 | 正向 | 当前 | 提高企业为自动化试点付费的意愿 | 需要证明 Rhoda 能以经济方式降低对人工依赖 |
| 物流中大量人工任务 | 正向 | 当前 | 支撑自动化用例的长跑道 | 需要工作流层面的渗透率证据 |
| 2026 年 AI 预算扩张 | 正向 | 当前 | 让实验性物理 AI 预算项更容易被批准 | 需要具名企业买方,而不只是调研意向 |
| 软件定义仓库趋势 | 正向 | 当前 | 让编排和智能层支出更容易解释 | 需要证据说明 Rhoda 能接入现有 WES/WMS 栈 |
| RaaS 与服务模式 | 正向 | 近期 | 帮助买方不用一次性巨额投入也能吸收新软件 | 需要 Rhoda 披露商业模式 |
| 资本成本和集成负担 | 负向 | 当前 | 即便技术演示强,也会拖慢采购 | 需要 ROI 和实施周期数据 |
| 关税 / 宏观不确定性 | 负向 | 2025-2026 | 推迟新建项目,并抬高设备成本 | 需要证明 Rhoda 能在存量改造中拿单 |
| 移动自动化预测下调 | 负向 | 当前 | 显示物理 AI 热情在不同子赛道并不均匀 | 需要分赛道验证需求 |
| 缺少收入 / 定价披露 | 负向 | 当前 | 阻碍自下而上的 SOM 测算 | 需要披露定价、ARR 和客户数量 |
方向反映这些因素对 Rhoda 式软件采用的可能影响。本表同时保留顺风和摩擦,而不假设物理 AI 会线性普及。
[CM023, CM024, CM026, CM028, CM030, CM031]2.5 图表
03竞争格局
3.1 竞争版图与买方替代方案
Rhoda AI 处在一个拥挤但仍在成形的物理 AI 市场,买方可以用几种很不同的方法解决同一件事。第一类是 Rhoda、Skild AI、Physical Intelligence 和 FieldAI 这样的中立机器人脑供应商。它们承诺的是横向软件杠杆:一个智能层跨越多种机器人、任务和环境。Rhoda 属于这一类,因为 FutureVision 被明确包装为一个智能层,预计授权给不同机器人硬件和软件平台,而不是锁在一台自家机器上。Skild 提出类似的跨形态论点,FieldAI 称 EDGE 是“跨机器人共用一个大脑”,Physical Intelligence 称 π0 是能控制不同机器人的通用机器人策略。 第二类是既有平台方:NVIDIA Isaac GR00T 和 Google DeepMind Gemini Robotics。它们不需要靠机器人模型收入来证明投资合理。NVIDIA 能用 GPU、模拟器和推理需求补贴开放模型;DeepMind 能把机器人纳入 Gemini 更大的基础模型栈和合作伙伴网络。第三类是 Figure 和 Apptronik 这样的垂直整合人形机器人厂商,把自研模型、一个机器人家族和制造路线图捆在一起。第四类是 Dexterity 这样的生产专精厂商,以及更相邻的仓库 AI 既有方如 Covariant,范围往往更窄,但在单一工作流里部署证明更强。对 Rhoda 来说,核心竞争问题不是物理 AI 有没有价值,而是视频优先的中立大脑能否在平台方和垂直玩家压缩市场之前,赢得足够部署数据。[CP001, CP003, CP006, CP015, CP018, CP022]
| 公司 | 类别 | 最新公开规模 | 主要目标 | 数据策略核心 | 部署 / 商业模式 | 相比 Rhoda 的关键短板 |
|---|---|---|---|---|---|---|
| Rhoda AI | 直接同类 / 中立大脑 | $450M Series A 轮;估值 $1.7B | 制造和物流中的工业操作 | 互联网规模视频预训练 + 10–20h 机器人后训练 | 未来跨硬件和软件平台做软件授权 | 还没有独立基准或大规模公开部署集 |
| Skild AI | 直接同类 / 中立大脑 | $1.4B 融资;估值 >$14B | 跨本体通用机器人智能 | 仿真 + 互联网视频 + 遥操作 + 部署反馈 | 软件大脑,通过伙伴 / OEM 分发 | 基准披露少于估值和营销暗示 |
| Physical Intelligence | 直接同类 / 研究型模型实验室 | 开放技术披露;openpi 生态 | 通用机器人控制和灵巧操作 | VLM 预训练 + 多机器人灵巧数据集 | 开源 / 社区 + 未来企业层 | 工业渠道杠杆不如 Rhoda 授权逻辑明确 |
| Figure AI | 相邻对手 / 垂直人形机器人 | 估值 $39B;BMW 试点 | 制造、物流和最终家庭场景中的人形劳动力 | 人类视频 + Helix 自有机队数据 | 硬件销售 + 服务 + 工厂规模 | 绑定 Figure 硬件,执行资本开支重 |
| Dexterity | 生产专精厂商 | 宣称生产中完成 100M 次自主动作 | 仓库和物流运营 | 已部署仓库系统的生产动作日志 | 全班次仓库部署 | 工作流范围窄于 Rhoda 的通用叙事 |
| FieldAI | 相邻同类 / 工业自主 | 横跨三大洲部署 | 建筑、工业、能源和现场作业 | 信念世界模型 + 风险感知自主 + 部署数据 | 面向工业机器人机队的软件智能 | 对双臂工厂操作的聚焦没那么直接 |
| NVIDIA GR00T | 平台在位者 | 开放模型 + 仿真器 / 计算栈 | 人形机器人 OEM 和机器人开发者 | Human EgoScale 视频 + 机器人演示 | 用模型访问拉动 NVIDIA 生态采用 | 经济激励更偏向平台锁定,而不是中立软件经济性 |
| Google DeepMind Gemini Robotics | 平台在位者 | Gemini 2.0 机器人项目,已有可信测试者 | 通用机器人助手和 OEM 伙伴 | Gemini 基础模型 + 机器人微调 | 模型 / API 生态 + 伙伴网络 | 商业包装和定价仍不透明 |
公开规模信号混合了融资、估值和已披露部署,因为该品类的标价和收入大多不透明。
[CP001, CP002, CP004, CP006, CP010, CP013]Rhoda 与七个主要替代方案的序数定位。横轴是硬件无关性(越高表示跨机器人越中立)。纵轴是已披露部署密度(越高表示公开生产证明越多)。
坐标轴是有证据支撑的序数评分,不是已发布的数值基准。
[CP001, CP004, CP008, CP011, CP014, CP016]3.2 直接同业、相邻对手与平台威胁
Skild AI 是 Rhoda 在 2026 年最明显的直接压力点,因为它用多得多的资本追求同一个中立大脑逻辑。Skild 的公开叙事结合了估值、伙伴叙事和通用控制野心:2026 年 1 月融资 $1.4 billion,估值超过 $14 billion,并坚持认为跨多种具身形态共享一个模型才是唯一可扩展答案。Physical Intelligence 是最接近的技术同业。π0 被公开记录为一个视觉-语言-动作流匹配模型,建立在预训练 VLM 和来自八种机器人的数据之上;openpi 让开发者和研究者能看清这套栈。Figure 是另一类对手:不是中立软件,而是垂直整合的人形机器人供应商,Helix 模型、Figure 03 硬件和 BotQ 工厂构成同一个逻辑。Dexterity 和 FieldAI 重要,是因为它们说明,即使更宽的“通用型”市场仍未验证,买方仍会奖励仓储和工业现场里窄而生产级的物理 AI。 NVIDIA 和 Google 是 Rhoda 不能忽视的既有平台方。GR00T N1.7 开放、可商业授权,并由 NVIDIA 的模拟器、工具链和算力生态支撑。Gemini Robotics 明确偏 VLA,但 Google 声称它比早期模型有更强基准测试泛化表现和更广具身支持,并有覆盖 Apptronik、Agility 和 Boston Dynamics 的可信测试者网络。Apptronik 本身与其说是直接模型实验室对手,不如说是战略信号:如果 OEM 更偏好垂直整合人形机器人加 Gemini 捆绑包,中立授权层就会被挤压。Covariant 更远,但仍是相关先例,因为它代表仓库 AI 路线:工作流深度,而不是硬件广度。不过公开层面,它目前官方资料面比新一代基础模型实验室薄得多,因此更像先例案例,而不是首要基准。[CP004, CP005, CP006, CP008, CP010, CP011]
| 购买标准 | Rhoda | Skild | Physical Intelligence | Figure | Dexterity | FieldAI | GR00T / Gemini |
|---|---|---|---|---|---|---|---|
| 核心策略架构 | 因果视频预测 + 逆向动力学 | 分层通用机器人大脑 | VLA 流匹配 | 人形机器人 VLA | 仓库物理 AI 智能体 | 信念世界模型 | 开放 / 平台 VLA |
| 本体立场 | 硬件无关授权 | 机器人无关 | 跨机器人通用控制 | 单一 Figure 机器人家族 | 面向特定任务的系统 | 一个大脑适配多类机器 | 多本体,但由生态主导 |
| 最强公开证据 | 生产式演示和试点说法 | 伙伴叙事 + 部署 | 已发布任务对比,基准为 OpenVLA/Octo | BMW 试点和出货叙事 | 全班次仓库作业 | 多大洲工业部署 | 基准和平台公告 |
| 数据护城河基础 | 网络视频 + 本体后训练 | 规模和部署飞轮 | 开放 + 自有多机器人数据 | 机队 + 人类视频 + 硬件遥测 | 生产中的自主动作 | 工业现场部署 | 平台规模的人类 + 机器人数据 |
| 开放性 | 封闭自研栈 | 封闭企业平台 | openpi 公开仓库 | 封闭自研 | 封闭自研 | 封闭自研 | 模型 / 工具层更开放 |
| 分发杠杆 | 早期,由伙伴驱动 | 增长中的 OEM / 工厂伙伴 | 研究 / 开发者生态 | 垂直硬件渠道 | 仓库运营商关系 | 工业伙伴网络 | 计算、仿真和 AI 平台分发 |
单元格汇总截至 runDate 公开检索到的最强证据,并刻意标注商业化方式,而不是强行填补缺失的价格披露。
[CP005, CP008, CP013, CP016, CP019, CP022]矩阵强调战略能力广度,而非纯模型质量:按竞争者类别比较数据广度、开放度、分发杠杆和部署证据。
高 / 中 / 低是基于公开证据和包装方式的分析师判断,不是供应商评级。
[CP008, CP009, CP019, CP021, CP023, CP028]3.3 数据策略、部署模式与定价压力
比较 Rhoda 与同业,最干净的轴是数据策略。Rhoda 的 DVA 栈押注互联网规模视频加少量特定具身机器人数据。这不同于 Skild 的四来源叙事:仿真、互联网视频、远程操作和部署反馈;也不同于 Physical Intelligence 的多机器人灵巧操作数据集加 VLM 预训练;不同于 Dexterity 在生产中完成 100 million 次自主动作的说法;也不同于 NVIDIA 建在 20,000 hours 人类 EgoScale 视频加机器人示范之上的 VLA 配方。Google DeepMind 也更靠近 VLA 一侧,即便它现在强调基准测试泛化和具身推理。换句话说,Rhoda 的独特性不只是“使用视频”——许多对手也用——而是因果视频预测是策略核心,而不是辅助数据来源或模型组件。 商业模式差异同样大。Rhoda 和 Skild 暗示软件授权。Physical Intelligence 把开源开发者触达与未来企业层级混合。NVIDIA 免费提供或开放模型层重要部分,用来销售周边栈。Figure 和 Apptronik 靠完整机器人、服务和工厂规模变现。Dexterity 向仓库工作流销售生产系统,而整个领域几乎没有公开标价。这意味着买方常常会根据部署风险、支持负担和渠道力量选择,而不是根据价格透明度选择。Rhoda 的硬件无关立场,在客户已经拥有机器人或想保留供应商灵活性时很有吸引力,但也意味着 Rhoda 必须解决垂直供应商可藏进单一软硬件合同里的难啃集成工作。[CP002, CP007, CP013, CP018, CP020, CP025]
| 公司 | 商业打包 | 硬件立场 | 公开定价可见度 | 变现对象 | 对买方的含义 |
|---|---|---|---|---|---|
| Rhoda | 授权逻辑 / 伙伴平台 | 对机器人硬件和软件保持中立 | Unknown | 智能层和部署支持 | 如果集成跑通,对既有机器人机队有吸引力 |
| Skild | 企业软件 / 伙伴合作 | 跨本体中立 | Unknown | 通用机器人大脑和部署服务 | 渠道伙伴关系比标价更重要 |
| Physical Intelligence | 开源 + 未来企业层 | 跨机器人中立 | Unknown | 模型、支持和自有数据优势 | 试验成本更低,但商品化风险更高 |
| Figure | 机器人 + 服务 + 工厂规模 | 自有人形机器人硬件 | 未公开 | 机器人、软件和运营 | 单一负责供应商,但硬件灵活性低得多 |
| Dexterity | 按工作流部署 | 围绕仓库系统调优 | 未公开 | 生产自动化结果 | 在可靠性比开放性更重要的场景能赢 |
| Apptronik | RaaS 模式 | 自有人形机器人硬件 | 未公开 | 机器人即服务 | 在替代人工用例中,可替代中立软件 |
| NVIDIA / Google | 平台 / 生态 | 参考硬件 + 伙伴机器人 | 模型定价不透明 | 计算、仿真、API 和生态锁定 | 从采用总成本看,可能压低独立软件层的空间 |
多数公司披露打包方向,但不披露标价,因此本表比较变现界面,而不是不可得的合同金额。
[CP001, CP018, CP030, CP037]| Rhoda 护城河主张 | 威胁 | 主要压力来源 | 严重度 | 为何重要 | 尽调问题 |
|---|---|---|---|---|---|
| 视频优先的因果模型在结构上不同 | 架构模仿 | Skild、PI、世界模型实验室 | 高 | 单靠模型架构很少能长期保持专有 | 要求提供消融数据,说明 DVA 相比 VLA 基线独特带来了什么 |
| 机器人数据需求低 | 基准赶超 | Google、NVIDIA、PI | 高 | 对手已经发布更明确的对比基准材料 | 获取与具名基线的并排任务和失败率对比 |
| 硬件无关授权 | 平台捆绑 | NVIDIA 和 Google | 高 | 既有厂商可把模型与算力、仿真或 AI 技术栈经济性打包 | 摸清客户愿意为中立性支付多少,以及与平台打包折扣如何取舍 |
| 封闭自研技术栈 | 开源商品化 | openpi、开放模型 | 中 | 开发者可能在别处做原型,只为部署增量付费 | 厘清哪些数据、工具或支持仍属自研且不可替代 |
| 工业任务聚焦 | 垂直专精厂商执行更快 | Dexterity, FieldAI | 中 | 通用厂商铺开前,专精厂商可能先拿下生产工作流 | 量化 Rhoda 在已有专精厂商生产案例的工作流中的胜率 |
| 早期部署飞轮 | 资本与渠道劣势 | Skild、Figure、平台型既有厂商 | 高 | 资金更足的对手能更快买到分销、试点和硬件资源 | 审核客户管线、试点转化和扩张速度假设 |
严重度是基于已检索证据的分析师判断,而不是公司提供的风险评分。
[CP027, CP028, CP029, CP030, CP031, CP034]3.4 差异化耐久性与竞争压力
Rhoda 的差异化可信,但仅凭公开证据还不算耐久。核心逻辑——因果视频预测应比重 VLA 的替代方案更能泛化、也更少依赖机器人数据——是自洽的,并且与许多同业主导的语言优先框架有实质差异。但在这个品类里,架构不是持久护城河。平台公司能模仿接口模式,开源开发者能复现栈的部分组件,买方最终在意的是错误恢复、部署吞吐和可支持性。因此,Rhoda 必须比资本更充足的对手更快把视频优先的先发优势转成数据和部署飞轮。Skild 在融资和公开伙伴叙事上已经更靠前。Physical Intelligence 在公开技术披露上更靠前。Figure 在已披露工业试点证明上更靠前。NVIDIA 和 Google 在生态杠杆上更靠前。 持久压力点在分销、基准测试可见度和切换成本。Rhoda 一旦嵌入客户工作流,集成复杂度应会形成有意义的锁定效应。在此之前,市场要软得多。如果中立大脑市场商品化,NVIDIA 和 Google 可以靠平台捆绑压低定价;如果客户偏好一个可问责供应商,Figure、Apptronik 和 Dexterity 可以凭全栈可靠性赢。公开定价不透明、缺少标准化跨公司基准测试,也让当前市场异常依赖叙事。对投资人来说,正确问题不是“Rhoda 的方法技术上是否有趣?”,而是“Rhoda 能否在既有巨头和垂直玩家把它挤进狭窄位置前,把技术独特性转成部署密度?”[CP027, CP028, CP029, CP030, CP031, CP034]
紧凑汇总与 Rhoda 中立“大脑”逻辑最相关的竞争压力。
[CP011, CP030, CP031, CP032, CP039, CP040]3.5 图表
04财务情况
4.1 收入模式与商业表面
Rhoda 的公开材料显示,变现模式比纯软件授权或纯机器人销售都更宽。FutureVision 被描述为一个智能层,今天驱动 Rhoda 系统,未来预计授权给合作方硬件和软件平台。这个表述指向软件平台野心。但公司同时推广自己的机器人平台和公开任务演示,意味着部署服务、系统集成,以及至少一部分内部硬件敞口。落实到承销,Rhoda 看起来是一家混合型前沿机器人公司:想捕获类似软件的杠杆,但还没有摆脱类似硬件和服务的执行要求。 商业证据点仍处早期。官网和发布稿指向汽车、制造、物流和电商工作流,再加上一个制造基准测试:据称在无人干预下,用不到两分钟完成一个组件处理周期。它有方向性价值,因为叙事不再只停留在实验室演示。但同一套公开材料仍缺少具名客户、客户数、收入、ARR、定价、ACV 和合同结构。所以商业表面已经可见,变现表面仍不透明。现有证据支持的是“存在试点需求和可信工业兴趣”,而不是“Rhoda 已经降低收入质量风险”。[CI001, CI002, CI003, CI004, CI006, CI007]
| 收入来源 | 机制 | 计费单位 / 合同 | 当前状态 | 收入质量 | 尽调事项 |
|---|---|---|---|---|---|
| FutureVision 平台授权 | 向 Rhoda 或伙伴系统授权机器人智能层 | 未知;可能是企业软件 / 平台合同 | 已有公开描述,但未量化商业收入 | 公开层面尚未验证 | 获取 MSA 样本、价格表和续约机制 |
| 试点部署 | 工业客户试点和部署项目 | 按项目或试点里程碑计费 | 资金用途中明确提及 | 早期形态,规模化前可能不是经常性收入 | 索取活跃试点清单、成功标准和转化率 |
| 部署 / 集成支持 | 系统启动、工作流梳理、现场集成、安全工作 | 服务或实施费(未披露) | 由运营模式和招聘结构推断 | 可能有意义,但不透明 | 把服务收入与经常性软件收入拆开 |
| Rhoda 运营系统 / 硬件相邻收入 | 自有机器人平台或整套系统部署 | Unknown | 产品定位有所暗示,但商业细节未披露 | 规模较大时可能稀释软件毛利 | 厘清硬件收入占比、COGS 和资产所有权 |
| OEM / 伙伴长期授权 | FutureVision 嵌入第三方硬件或软件栈 | 未知的多年授权结构 | 更像战略愿景,而非已披露的当前业务 | 上行空间大,但尚未验证 | 识别首个 OEM 交易及其单位经济性 |
概括公开信息暗示的收入架构;所有商业条款仍未披露。
[CI001, CI002, CI003, CI004, CI009, CI014]| 价格 / 合同要素 | 公开披露值 | 公开记录实际说了什么 | 置信度 | 重要性 | 尽调事项 |
|---|---|---|---|---|---|
| 标价 / 订阅费 | null | 检索到的来源均未公布价格、ACV 或按席位 / 机器人计费 | 低 | 决定软件可扩展性和客户采用阻力 | 获取费率卡或已签客户报价 |
| 实施费 | null | 公开资料没有服务定价或部署费用表述 | 低 | 实施占比高会压低毛利率 | 将实施费与经常性软件收入拆开 |
| 硬件价格 / 租赁条款 | null | Rhoda 推广自有平台,但不公布商业条款 | 低 | 需要用来判断硬件敞口和营运资本占用 | 索取 BOM、毛利率和销售条款 |
| 试点转生产条款 | null | 提到试点,但未披露付款结构 | 低 | 试点占比高的收入未必能转成经常性 ARR | 索取转化漏斗和试点合同模板 |
| 伙伴授权经济性 | null | FutureVision 授权停留在战略描述,没有数字 | 低 | 决定 Rhoda 能否赚到类似软件的经济性 | 索取首份伙伴协议或条款清单 |
null 代表未披露,不代表为零;核心问题是公开变现透明度缺失。
[CI001, CI007, CI009, CI037]如果模型可迁移,Rhoda 公开商业化故事必须把工作流转化为经常性财务产出。
这是一座逻辑桥,不是量化瀑布图;公开定价和 ACV 不可得。
[CI001, CI003, CI004, CI009, CI014]4.2 GTM 成熟度与收入质量
公开 GTM 证据在 Rhoda 谈部署时最强,在投资人需要合同经济性时最弱。公司似乎通过直接试点工作和未来授权给合作方软硬件平台的组合,触达工业运营商。信息口径偏企业、偏集成,符合 Rhoda 正在追逐的领域。但公开记录没有说明交易究竟围绕试点里程碑、经常性软件订阅、捆绑硬件加服务套餐,还是长期 OEM 授权来设计。这个缺席很关键,因为每种收入模式都意味着完全不同的利润率画像和营运资本需求。 招聘信号让商业化姿态看起来真实,但仍处于规模化前。公开 Ashby 招聘板显示 33 个开放职位,全部在 Palo Alto,明显集中在研究和软件,同时定向招聘硬件、供应链和运营岗位。这更像一家公司在为更深产品化配人,而不是已经建成成熟的多站点销售和现场服务机器。正确解读是:Rhoda 正在投资走向商业化,不是已经证明可规模化、可重复收入。在客户集中度、合同期限和续约数据披露前,公开的收入质量逻辑仍由叙事主导。[CI006, CI007, CI008, CI010, CI011, CI014]
| 指标 | 公开值 / 代理指标 | 置信度 | 重要性 | 尽调事项 |
|---|---|---|---|---|
| 收入 / ARR | null | 低 | 没有收入规模,外部投资者无法锚定倍数或销售效率 | 索取月度经常性 / 非经常性收入桥表 |
| 毛利率 | null | 低 | 需要拆出软件杠杆与部署拖累 | 索取软件、服务和硬件毛利率 |
| CAC / 回本周期 | null | 低 | 企业机器人销售周期可能又长又贵 | 提供销售漏斗、胜率和分队列回本周期 |
| 客户留存 / NRR | null | 低 | 没有续约数据,收入耐久性未知 | 提供队列留存和扩张指标 |
| 招聘强度 | 33 个开放岗位;集中在研究和软件 | 中 | 暗示在公开收入验证前,运营费用先上台阶 | 将开放岗位与当前在岗人数和薪资预算对齐 |
| 成本结构组合 | 算力 + 软件基础设施 + 硬件工程 + 部署支持 | 中 | 帮助判断 Rhoda 最终能否呈现软件型经济性 | 提供按职能拆分的预算和预测 |
本表刻意强调缺失项,因为公开的单位经济性披露实际上缺位。
[CI006, CI007, CI010, CI011, CI012, CI024]把 Rhoda 公开信息里最大的成本驱动因素、缺失的商业输入,与仍未解决的单位经济问题连起来。
由于收入、定价、毛利率、CAC 和留存均未披露,本图基于定性公开信息搭建。
[CI007, CI012, CI024, CI027, CI028, CI037]4.3 成本结构与资本强度
Rhoda 的成本结构几乎肯定比传统 AI 应用公司更重。研究说明清楚表明,DVA 依赖互联网规模视频预训练、长上下文视频记忆和自回归生成,这些都意味着大量算力、存储、数据工程和模型运营成本。与此同时,公司在物理侧显然不是轻资产:官网宣传带定制执行器和安全等级视觉的 Rhoda 机器人平台,在线职位包括 VP of Hardware,以及供应链和集成岗位。这个组合指向混合支出画像:前沿模型训练,加上物理系统工程和部署。 反方论点是,如果技术逻辑跑通,Rhoda 仍可能实现有吸引力的单位经济性。公司称部分任务可用约 10 hours 机器人数据学会;如果得到验证,相比需要大得多的任务特定数据集的机器人栈,远程操作成本可能大幅下降。如果 FutureVision 确实能跨多个硬件合作方迁移,公司也可能在早期偏硬件的建设期之上赚到类似软件的杠杆。但公开数据还不能让外部人判断两种未来哪一种会发生。现有记录足以得出“现在资本密集,之后可能像软件”的结论,不足以量化这个转变何时、是否会发生。[CI002, CI011, CI012, CI013, CI027, CI028]
定性地图,展示在公开收入证明出现前,最可能消耗 Rhoda 大额 Series A 资金的成本中心。
渲染为流程图,因为公开披露只支撑方向性成本桶,不支撑预算分配的数值矩阵。
[CI005, CI011, CI012, CI026, CI027, CI028]4.4 资本充足性与融资依赖
Rhoda 的好消息是,绝对融资额大到足以产生影响。首个披露轮次 $450M,让管理层一开始的资产负债表深度就超过多数机器人创业公司多轮融资之后的水平。官方材料和法律报道也用偏增长的方式描述募资用途:工程投入、工业部署、客户试点和团队增长。这说明资本被定位成扩张燃料,而不是为既有义务再融资。 坏消息是,公开数据仍太薄,无法把轮次规模换算成现金跑道。烧钱速度和账上现金都未披露。检索到的来源没有显示债务工具或项目融资结构,SEC 公司搜索也没有 Rhoda AI 或 Rhoda Ai Corporation 名下的公开发行人记录,但这种缺席并不能解决承销问题。一家公司仍可能通过模型训练、招聘和部署支持快速消耗资本,而无需公开披露。因此,本轮融资降低了短期破产风险,却没有消除融资依赖这一尽调主题。如果部署和收入转化滞后,而算力和招聘支出像招聘板暗示的那样扩大,Rhoda 在达到持久现金生成前,仍可能需要更多大额轮次。[CI005, CI015, CI017, CI018, CI019, CI021]
| 项目 | 公开状态 | 置信度 | 影响 | 尽调事项 |
|---|---|---|---|---|
| 最新披露融资 | 2026-03-10 宣布 $450M Series A 轮 | 高 | 刚公开亮相的机器人创业公司,短期资本底座很强 | 确认总融资额与净到账、交割时间表 |
| 估值 | 二级来源报道 ~$1.7B | 中 | 高估值抬高了下一轮或流动性事件前的执行门槛 | 核对已签融资文件中的估值 |
| 账上现金 | null | 低 | 公开数据无法估算现金跑道 | 索取当前资产负债表和月度现金报告 |
| 月度烧钱 | null | 低 | 支出未知,外部无法量化融资依赖 | 提供按职能拆分的烧钱速度和情景预测 |
| 现金跑道(月) | null | 低 | 没有烧钱和现金数据,任何现金跑道估计都是编造 | 提供基准和下行情景下的现金跑道 |
| 资金用途 | 研发、工程、部署、试点、团队扩张 | 高 | 资本投向规模化,而非偿债 | 将资金用途映射到季度预算 |
| 债务 / 项目融资义务 | 未找到公开披露 | 低 | 杠杆可能为零,也可能只是未披露 | 确认债务、租赁、担保和契约条款 |
| 下一轮触发条件 | null | 低 | 公开记录看不出下一轮融资取决于 ARR、试点还是硬件规模 | 索取董事会运营计划和融资里程碑 |
本表以融资时间线作背景;此处用到的事实只形成财务情况章节的本地声明。
[CI005, CI015, CI016, CI017, CI018, CI019]只使用可支撑的公开边界;烧钱和现金跑道未披露时,区间有意收敛为「公开无法衡量」。
公开记录能支撑已披露融资金额的精确数值;对收入 / 定价 / 客户指标,只能支撑披露数量为 0 的区间,不能支撑烧钱或现金跑道的数值估计。
[CI006, CI007, CI008, CI015, CI016]4.5 财务结论与尽调阻塞点
Rhoda 的公开财务画像只有在尽调愿意先于收入证明承销技术承诺和投资方质量时,才具备可投性。乐观情景足够清晰:很大的轮次、强投资方阵容、可信研究背景,以及至少看起来具备商业方向的具体工业工作流。悲观情景同样清晰:没有披露收入、定价、具名客户、毛利率、烧钱速度、现金余额,也没有独立验证的部署经济性。这正是能让一个有吸引力的物理 AI 叙事跑在公开承销纪律前面的披露模式。 检索集中最强的怀疑来源 robotics.press 认为,估值由未来潜力而非市场证明支撑。这个判断可能略微过度,但抓住了核心问题。严肃尽调需要客户访谈、合同结构、定价、集中度、烧钱、现金跑道和利润率数据;没有这些,就很难把机会作为财务投资来承销,只能作为物理 AI 主题下注。在这些输入私下披露前,唯一诚实的公开结论是:Rhoda 的资本基础和技术野心很强,但收入质量和利润率路径仍未证明。[CI006, CI007, CI008, CI025, CI029, CI031]
| 缺失的私有指标 | 重要性 | 当前公开替代信息 | 对投资判断的影响 | 具体尽调路径 |
|---|---|---|---|---|
| 收入 / ARR | 估值和 GTM 质量的核心锚 | 只有试点 / 部署叙事,加上一项公司自述的制造 KPI | 无法用基本面检验估值 | 获取月度收入桥表和最新 ARR |
| 定价 / ACV / 合同结构 | 建模收入质量和利润率必须有 | 完全没有公开定价 | 无法区分软件平台与服务占比较高的业务 | 审阅已执行客户合同和订单 |
| 客户集中度 | 需要评估依赖风险和收入耐久性 | 未披露具名客户或数量 | 少数试点就可能解释全部当前牵引力 | 索取前 10 大客户收入组合 |
| 按收入流拆分的毛利率 | 拆出软件杠杆与硬件 / 服务拖累 | 没有毛利率披露 | 无法建模盈利路径 | 提供软件、服务和硬件毛利率 |
| 烧钱速度和现金余额 | 现金跑道和融资依赖分析必须有 | 大额融资已知,烧钱速度未知 | 现金跑道判断只能是猜测 | 审阅现金瀑布和 12 个月计划 |
| 留存 / 续约 / NRR | 需要用来判断经常性经济性 | 没有续约或队列披露 | 无法判断试点能否转成耐久 ARR | 提供队列留存和续约历史 |
每一行都是公开投资判断的真实阻断点,不是风格上的「有了更好」。
[CI006, CI007, CI008, CI025, CI029, CI037]05产品与技术
5.1 产品定义与 DVA 架构
Rhoda 已交付的产品不只是机器人演示,也不只是研究模型;它是一个名为 FutureVision 的拟议智能层。用客户语言说,FutureVision 计划位于传感器和执行之间,持续观察世界,把未来状态预测成视频,再足够快地把这些预测转成动作,用于真实世界控制。这个架构重要,因为 Rhoda 正在反对一种主流框定:机器人基础模型主要是视觉-语言-动作系统。在 Rhoda 的表述里,语言仍可给模型加条件,但重心是因果视频预测:世界随时间运动,所以控制策略应该直接从视频学习运动、物理和物理交互,而不是主要依赖远程操作机器人轨迹。 官方研究博客把架构讲得异常具体。DVA 从一个从零训练、基于互联网规模视频的因果视频模型开始。Rhoda 不只是在编码完整序列后预测少数几帧,而是使用一种名为 Context Amortization 的训练方法,在一段干净长历史上下文的每个位置预测未来帧。运行时,KV-caching 复用已编码上下文,系统不必为长时域条件反复支付完整算力成本。预测出的未来随后交给一个独立逆动力学模型,由它推断实现想象结果所需的精确末端执行器运动。Rhoda 的 Leapfrog Inference 把正在执行的动作和下一轮预测周期重叠起来,让模型思考时系统仍在移动。这是一套具体且自洽的栈,也是本章的关键技术差异点。[CE001, CE002, CE003, CE004, CE005, CE006]
| 模块 / 资产 | 主要用户 | 状态 / 成熟度 | 关键差异化 | 尽调缺口 |
|---|---|---|---|---|
| FutureVision | 工业客户 / OEM 伙伴 | 商业平台逻辑;2026 年 3 月公开发布 | 硬件无关的智能层,而不是固定机器人应用 | 没有公开定价或伙伴集成参考架构 |
| DVA 因果视频骨干 | Rhoda 研究与部署团队 | 有研究支撑;用于公开演示 | 视频预测是策略核心,而非辅助规划器 | 没有针对具名 VLA 基线的第三方基准测试 |
| 逆动力学转换器 | 具身集成团队 | 已在演示中运行;小模型适配器 | 用少量特定具身数据集,把预测未来映射成动作 | 没有已发布的延迟和控制频率表 |
| 长上下文记忆 / 上下文内演示模式 | 工作流设计者 / 操作员 | 在研究演示中展示 | 数百帧视觉上下文支持一次性模仿和端到端任务记忆 | 跨更多任务和客户的泛化仍未验证 |
| Rhoda 机器人平台 | 现场操作员 / 集成商 | 已公开展示硬件能力 | 25kg 额定载荷、40kg 峰值、安全级视觉、执行器制动 | 没有详细 BOM、维护计划或认证文件 |
| 评估与推演工具 | 研究 / 可靠性工程师 | 由自回归视频推演和基础设施招聘推断 | 可调试的视频推演,加上用于数据采集和模型评估的云工作流 | 没有公开可观测性工具或客户事故案例 |
成熟度评级反映公开发布和演示证据,而非已披露的内部 TRL 框架。
[CE001, CE002, CE007, CE015, CE017, CE020]| 层 / 流程 | 作用 | 依赖 | 主要上行 | 风险 |
|---|---|---|---|---|
| 网页视频预训练 | 学习运动和物理先验 | 大规模通用视频数据 | 以低成本把训练多样性扩到机器人演示之外 | 数据来源和版权敞口未知 |
| 因果视频模型 | 预测未来视觉状态 | 长上下文和高效训练目标 | 把动力学放在控制核心 | 如果缓存和重叠做得不细,算力开销可能很高 |
| 上下文摊销 | 在每个序列位置训练未来预测 | 干净的长上下文窗口 | 让数百帧训练可跑通 | 公开证据没有量化绝对算力成本 |
| KV 缓存推理 | 在步骤之间复用已编码上下文 | 稳定的长上下文运行路径 | 减少冗余计算 | 延迟数字未公开披露 |
| Leapfrog 推理 | 将动作执行与下一次预测重叠 | 以动作为条件的未来重叠 | 即使存在推理延迟,也能支撑连续控制 | 未公开相对更简单循环的延迟或加加速度指标 |
| 逆动力学 | 把预测出的未来转成末端执行器动作 | 本体专属数据和转换模型 | 所需本体数据远少于完整策略重训 | 跨多种机器人类型的适配器表现未公开基准 |
| 闭环执行 | 观察 → 预测 → 执行 → 再观察 | 机器人传感器与执行器栈 | 可实时响应布局和物体变化 | 安全论证仍取决于未披露的底层控制与验证 |
本表把官方机制说法与已识别风险面放在一起;公开证据仍偏薄。
[CE002, CE005, CE006, CE007, CE008]五层堆栈,展示 Rhoda 如何把互联网级视频转化为已部署的机器人行为。
[CE001, CE002, CE005, CE006, CE007, CE008]5.2 数据效率、长上下文记忆与用例证据
Rhoda 最强的产品主张是数据效率。公司称,在因果视频模型已经从互联网规模视频吸收运动先验后,DVA 可以用大约 10–20 hours 机器人数据学会新的长时域工业任务。两个旗舰用例都刻意选择难啃场景:轴承拆袋转移和 Contico 容器拆解。据称,轴承拆袋转移只用了 11 hours 任务数据,却仍能处理断裂绑带、撕裂袋子、歪斜箱子等边缘情况,并连续运行 1.5 hours。容器拆解据称使用 17 hours 机器人数据,随后在处理重箱、部分可观测和随机碎片时连续运行 160 minutes。官方材料把两者都定位为生产式客户概念验证(PoC),而不是受控实验室基准测试。 长上下文视觉记忆是第二个定义性能力。Rhoda 明确把数百帧原生上下文,与许多只处理少数帧的 VLA 系统对比。猜杯游戏示例是这一能力的玩具基准测试:物体消失,杯壳移动,模型仍必须跨时间保持状态。更有商业相关性的是退货处理,Rhoda 称它能端到端运行,不需要手工设计的进度指示器或多阶段脚手架。单样本分类和绘图演示将同一思路从记忆延伸到上下文模仿:单次人类演示被注入上下文窗口,机器人在不更新权重的情况下模仿演示意图。这些机制和演示有吸引力,但仍是公司自写证据,而不是独立基准测试结果。[CE009, CE010, CE011, CE012, CE013, CE014]
| 用户任务 | 当前工作流 | Rhoda 解决方案 | 实测 / 声称收益 | 限制 |
|---|---|---|---|---|
| 轴承拆箱 | 人工拆包、倒箱和包装分拣 | 双臂 DVA 策略,具备长时程恢复行为 | 11h 机器人数据;声称 1.5h 自主运行 | 没有外部基准测试或单位经济性披露 |
| Contico 周转箱拆解 | 人工清理杂物、解锁并折叠箱体 | DVA 策略,具备重物推理和长上下文记忆 | 17h 机器人数据;声称连续运行 160 分钟 | 只有公司自写证据 |
| 退货处理 | 多步骤服装检查、折叠和重新包装工作流 | 无需工程化进度指示器的端到端长上下文策略 | 利用历史处理视觉相近状态中的歧义 | 没有跨站点或 SKU 的吞吐量分布 |
| 一次性分拣 | 人类演示目标 / 容器映射 | 把演示放入上下文窗口,用于上下文内模仿 | 声称无需更新模型权重即可单次学习 | 展示于演示环境,而非标准化客户基准测试 |
| 一次性绘图 | 人类演示目标形状和笔画顺序 | 基于上下文模仿最终形状和序列 | 迁移的是演示意图,而不只是运动轨迹 | 商业关联间接;主要是能力证明 |
收益来自公司披露的运行时长或数据效率说法,不应解读为独立审计过的 KPI 分布。
[CE010, CE011, CE014, CE015]从杂乱真实任务观察,到 DVA 预测、动作转译和持续恢复的运行流程。
[CE002, CE006, CE007, CE014, CE015]按模块展示能力成熟度。该图把技术上已演示的内容与商业上已验证的内容分开。
高 / 中 / 低 为分析师判断,用来反映已展示运行机制与经过外部验证的商业成熟度之间的差距。
[CE001, CE002, CE012, CE015, CE018, CE020]5.3 FutureVision 与重 VLA 路线的差异
概括 Rhoda 逻辑最锋利的说法是:许多 VLA 系统仍过于语言优先、上下文太短,无法弥合“真实世界鸿沟”。Rhoda 并不是唯一使用视频或互联网规模数据的公司,但比多数同业更激进,因为它把因果视频预测本身做成策略。系统先想象世界的下一段,之后才把想象出的未来转成动作。这不同于 GR00T、GR-2 和 Gemini Robotics;即便这些系统变得更多模态,或大规模使用人类视频,它们仍清楚处在 VLA 家族内。外部评论有助于解释 Rhoda 为什么认为这很重要。Mimic Robotics 认为,VLA 主干继承了语义,但没有继承物理动态,把昂贵的具身学习留给稀缺的机器人轨迹。Kempner Institute 也认为,互联网规模视频捕捉物理变换的方式,是图文预训练做不到的。 这不意味着 Rhoda 独占赛道。DreamGen 和其他世界模型努力说明,市场更多力量正在转向更丰富的视频预训练。NVIDIA 的 GR00T 路线更开放、更面向基准测试,Google 的 Gemini Robotics 通过 VLA 栈声称更强泛化指标和更广具身覆盖。换句话说,Rhoda 的逻辑有差异,但并不孤立。真正问题是,因果视频优先公式能否在生产中带来实质更好的稳健性、样本效率和运营调试。公开来源支持这个逻辑的可信度,尤其在上下文长度和可解释性上,但还没有用标准化第三方基线证明它优于同任务集上的具名 VLA 或世界模型。[CE016, CE026, CE027, CE028, CE029, CE030]
依赖图突出 Rhoda 在核心模型之外还需要什么,才能让 DVA 具备商业可信度。
[CE018, CE021, CE023, CE024, CE035, CE036]5.4 部署模式、开发者信号与信任缺口
Rhoda 的产品故事落在硬件无关的授权逻辑上。FutureVision 应先在 Rhoda 系统里起步,再扩展到合作方平台。这个策略有吸引力,因为 Rhoda 可能销售智能,而不用吸收作为完整机器人 OEM 的全部资本支出和制造负担。它也符合公司的招聘姿态。团队和职位页面显示出明显的全栈建设:硬件、世界模型、云基础设施、现场运营和模型训练都放在同一平台内。公司显然在投入数据采集、机队支持和模型迭代所需的机器;如果一个中立大脑供应商想支持异构部署,这些正是必需能力。 问题不是缺少技术细节,而是缺少独立验证。Coey 的批评是公平的:10 hours 数据并不是真正的承销问题。真正问题是在混乱的故障恢复、日志、监控和客户特定边缘情况中如何运营化。为本章审阅的来源中,没有找到公开第三方安全认证、正式模型审计或 DVA 标准化基准测试。官网宣传安全等级视觉、执行器刹车和三年可靠性主张,研究博客也认为视频推演有助于可解释性和安全行为检查。这些是积极信号,但仍是内部信号。同样,开发者信号目前更多来自招聘,而不是公共生态表面;审阅的官方材料没有找到公开 SDK、API 文档或代码仓库。这不会否定产品,但意味着外部证明栈仍比技术叙事暗示的更窄。[CE017, CE018, CE019, CE020, CE021, CE022]
| 控制 / 指标 | 目前公开状态 | 范围 | 缺口 |
|---|---|---|---|
| 安全级视觉 | 官网声称 | 机器人平台硬件 | 未公开认证材料或测试规程 |
| 每个执行器都有制动器 | 官网声称 | 机器人硬件故障安全姿态 | 未公开故障树或控制器审计 |
| 3 年连续运行说法 | 官网声称 | 可靠性 / 耐久性 | 未公开工作周期方法或现场队列 |
| 自回归视频推演 | 研究博客解释 | 可解释性 / 模型调试 | 没有证据显示监管方或客户认可其作为正式安全证据 |
| 闭环重规划 | 新闻和研究材料声称 | 运行鲁棒性 | 未发布标准化故障分类 |
| 正式安全认证 | 已审阅来源未发现 | 外部信任 / 采购 | 阻碍规模化工业投资判断的重大尽调项 |
| 公开安全 / 数据政策文档 | 已审阅来源未发现 | 客户保障与数据治理 | 本次审阅未发现公开 SDK、文档或明确的数据处理控制 |
各行区分公司声称的控制与外部验证的控制;已审阅来源中缺失,并不证明其不存在,但确实是尽调缺口。
[CE017, CE021, CE022, CE035, CE036]| 日期 / 阶段 | 功能或里程碑 | 状态 | 含义 | 来源 |
|---|---|---|---|---|
| Mar 2026 | 隐身 18 个月后公开发布 | 已完成 | Rhoda 从低调研发转向明确商业定位 | SE003 |
| Mar 2026 | 研究博客发布 DVA、逆动力学、上下文摊销和 Leapfrog 推理 | 已完成 | 作为刚公开的机器人初创公司,技术披露异常细 | SE002 |
| Mar 2026 | 高批量制造评估中单周期低于两分钟 | 已声称 | 显示叙事正从实验室演示转向工厂级 KPI | SE003 |
| 2026 | FutureVision 面向合作伙伴硬件 / 软件授权 | 规划 / 论点 | 硬件无关收入模式的核心 | SE003 |
| 2026 | Palo Alto 正在招聘基础设施与平台岗位 | 进行中 | 显示投入不止研究演示,还包括车队运营、训练和内部工具 | SE013 |
| 2026+ | 更广泛的安全、基准和生态证明 | 需要 | 要把技术新意转成采购信任,必须补齐 | SE011 |
最后一行是分析师对下一步必要事项的综合判断,不是公司发布的路线图项目。
[CE001, CE017, CE023, CE034, CE035]5.5 图表
06客户情况
6.1 公开客户表面与目标细分市场
Rhoda 的客户故事已经能看见,但更多是靠工作流而不是客户标识呈现。官网明确说,公司服务汽车、制造、物流和电商等行业客户,并用三个任务例子支撑:为物流客户处理退货、汽车装配线上的轴承倒料、在制造场景拆解 Contico。这足以判断其核心商业切入点。买方大概率不是消费者或小企业团队,而是想把可变物理作业自动化的企业运营或制造组织;传统脚本式机器人在这类工作上一直很难跑通。实际用户可能是工厂技师、机器人工程师、产线或仓库主管,付款方则是通过资本设备或自动化预算采购的工业企业。官方联系流程也偏顾问式,而非自助式,进一步印证其企业销售路径。同样重要的是缺失项。公开客户信息没有列出客户数、具名账户、合同结构,也没有给出宽泛工业垂类之外的地理分布,因此本章把 Rhoda 视为一家确有早期客户活动、但公开账户披露仍有限的公司。[CU001, CU002, CU003, CU004, CU005, CU006]
| 公开客群 | 买方 / 发起人 | 用户 | 付款方 | 工作流信号 | 关键缺口 |
|---|---|---|---|---|---|
| 汽车装配 | 制造工程 / 工厂自动化负责人 | 产线操作员和机器人工程师 | 汽车 OEM 或一级供应商 | 汽车装配线上的轴承卸箱 | 未披露具名账户或现场 |
| 制造业物料搬运 | 运营副总裁 / 持续改进负责人 | 工位操作员和维护技师 | 工厂运营方 | Contico 容器拆解和高批量零部件处理 | 仅公开一个量化 KPI |
| 物流退货处理 | 仓库运营 / 履约自动化负责人 | 仓库员工和主管 | 物流运营商或零售商 | 面向物流客户的端到端退货处理 | 未披露具名部署方或 ROI 指标 |
| 电商 / 全渠道履约 | 履约技术采购方 | 退货与分拣团队 | 零售商或 3PL | 官网将电商列为目标垂直行业 | 除退货处理外未公开其他工作流 |
各行反映公开的工作流证据,不是封闭式客户名单。公开披露给出垂直行业和任务,但没有披露客户数、地点或合同金额。
[CU001, CU003, CU004, CU005, CU008, CU024]展示 Rhoda 可能的企业客户旅程如何从工作流发现走向技术验证,最终进入规模化授权或站点扩张。
各阶段综合自公司关于客户试点、生产环境评估和未来授权计划的官方表述。公司未披露按账户划分的公开时间表。
[CU003, CU010, CU011, CU014, CU017]6.2 工作流证据真实存在,但账户证据仍不完整
积极解读是,Rhoda 的证据不止于光鲜的实验室视频。2026 年 3 月发布材料称,公司已在生产环境中展示自主运行,并量化了至少一次近期的大批量制造评估:无人工干预下,每个周期不到两分钟。更深入的技术研究文章进一步说明,Rhoda 表示两个客户任务样例是真实客户的概念验证,连续运行数小时无干预,其中一个倒料工作流使用 11 小时任务数据,一个容器拆解工作流使用 17 小时。这些都是有意义的信号,说明公司处理的是部署级可变性,而不只是适合刷榜的抓取任务。限制在于,公开记录仍未达到可供背书的客户证明。官方页面没有说出物流账户、汽车装配账户,或高产量制造评估背后的客户名称。第三方媒体补充了一些细节,包括有报道称其在一家超大型汽车工厂运行,但这些证据仍没有转化为具名部署方背书。按尽调口径,Rhoda 已有公开工作流证据和早期部署证据,但还没有公开账户证据。[CU009, CU010, CU011, CU012, CU013, CU014]
| 指标 | 公开值 | 日期 / 时效性 | 置信度 | 含义 | 缺失分母 |
|---|---|---|---|---|---|
| 具名公开客户 | 0 | 当前 | 高 | 有工作流证明,但没有账户证明 | 客户数和客户名单 |
| 已公开命名的垂直行业 | 4 | 当前 | 高 | Rhoda 并非围绕单一小众工作流定位 | 按垂直行业划分的收入结构 |
| 公开客户概念验证任务 | 2 | 当前 | 高 | 至少两个任务被呈现为真实客户 PoC | 还有多少未披露 PoC |
| 量化公开制造 KPI | 单周期 <2 min | 2026-03-10 | 高 | 至少有一个类生产 KPI 公开 | 基准周期时间和吞吐量分母 |
| 公开披露的客户试点 / 部署 | 提到扩张,但未给出具名数量 | 2026-03-10 | 中 | 商业推进看起来活跃 | 试点数量、部署站点数和转化率 |
本表区分可公开计数的内容与仍未披露的内容。null 或文本占位符表示缺少分母,而不是活动为零。
[CU006, CU009, CU010, CU014, CU017]| 公开客户标签 | 客群 | 部署 / 用例 | 生产 / 试点 | 结果 | 限制 |
|---|---|---|---|---|---|
| 未具名物流客户 | 物流 | 端到端退货处理 | PoC / 工作流证明 | 显示其能处理模糊的多步骤长上下文工作流 | 客户名称、站点数和 ROI 未披露 |
| 未具名汽车合作伙伴 | 汽车 | 装配线上从 10 kg 箱中卸出轴承 | PoC / 评估 | 官方和第三方证据称,该任务此前难以自动化 | 未披露具名 OEM、站点或稳态吞吐量 |
| 未具名制造合作伙伴 | 制造 | Contico 容器拆解 | PoC / 评估 | 研究文章称,采集 17 小时机器人数据后自主运行 160 分钟 | 未披露客户身份或节省人力数据 |
| 未具名高批量制造评估 | 制造 | 零部件处理工作流 | 生产环境评估 | 披露的评估中单周期低于两分钟,且无人干预 | 仅公开一个 KPI,客户仍未具名 |
枚举不完整,因为 Rhoda 披露的是工作流层面的证明,而非完整账户名册。各行总结不同公开证据点,不是完整部署清单。
[CU002, CU003, CU004, CU005, CU010, CU017]显示从宽泛垂直行业主张到客户级部署证明,公开可见度如何层层收缩。
计数按不同公开证明类别统计,而非内部 CRM 阶段。最后的 0 是有意保留,抓住了核心商业披露缺口。
[CU001, CU006, CU010, CU017]比较 Rhoda 可见工作流类别的公开证据质量,而不是按已点名账户比较。
评级是对公开证据深度的分析性概括。「强」表示公开记录包含具体运营主张;并不意味着该部署可作为独立客户背书。
[CU006, CU010, CU017, CU018, CU019, CU038]6.3 留存韧性、扩张和集中度需要直接尽调
公开材料能撑起一个先落地再扩张的叙事,但还不能证明它。Rhoda 表示融资将扩大工业部署和客户试点;Reuters 称该平台能适配大量机器人硬件,使制造商和物流运营商不用重建现有系统,也能部署智能机器人。这在商业上有吸引力,因为它降低买方迁移负担,并可能让 Rhoda 把智能层卖进现有机队,而不是只能销售捆绑的自有机器人。但缺失的留存数据很关键。已审阅的官方页面没有披露客户数、合同期限、续约、NRR、GRR、流失或满意度。制造评估或物流工作流也没有公开 ROI 或回本周期。因此,核心商业化问题不是 Rhoda 能否展示有意思的任务;而是这些任务能否跨站点重复,并跨预算周期留存。没有具名背书,公司还背着更高的集中度风险:少数试点密集的工业账户可能主导学习和收入,但市场目前无法衡量这项敞口。[CU011, CU012, CU013, CU014, CU015, CU024]
| 指标 | 公开值 | 证据质量 | 重要性 | 尽调要求 |
|---|---|---|---|---|
| 客户数 | 未公开披露 | 没有客户数,就无法判断集中度和先落地再扩张 | 提供当前客户数和活跃试点 | |
| 合同期限 / 续约节奏 | 未公开披露 | 续约结构决定试点能否沉淀为可持续收入 | 提供合同模板或续约历史样本 | |
| NRR / GRR / 流失 | 未公开披露 | 仅靠演示推不出收入耐久性 | 提供队列级留存指标 | |
| 客户愿意充当推荐人的程度 | 未公开披露 | 推荐意愿是外部最快检验部署质量的方法 | 安排至少两次客户推荐访谈 | |
| 运行持续性信号 | 两个 PoC 中均实现数小时无人干预 | 官方研究博客 | 显示早期任务持续性,但不能证明跨季度账户耐久性 | 展示干预率趋势和重复部署历史 |
null 值表示公开材料未披露指标,不代表绩效为零。最后一行抓住了公开材料中最强的耐久性替代信号:任务持续性,而非合同留存。
[CU017, CU018, CU019, CU024, CU025]| 扩张驱动因素 | 商业上行空间 | 集中度 / 摩擦风险 | 投资含义 |
|---|---|---|---|
| 向既有车队进行硬件无关授权 | 更大的 TAM 和更低客户切换成本 | 取决于第三方机器人硬件和集成质量 | 上行空间大,但合作伙伴执行很关键 |
| 从制造到物流的垂直跨度 | 多个工业切入点降低对单一工作流的依赖 | 公司可能仍只服务少数未具名工业客户 | 要判断是否多元化,需要账户级结构 |
| 可变任务中的工作流级证明 | 显示自动化能力有实质含量 | 证明仍未具名,因此难以做客户背调 | 可视为有希望,但还不能作为投资依据 |
| 融资资金留给工业部署和试点 | 为把评估转成重复部署留出空间 | 还没有多站点车队规模或转化率的公开证据 | 转化指标是头号尽调要求 |
本表关注同一组事实如何同时支撑上行空间和风险。公开部署数缺失时,应将集中度风险视为未解决,而不是忽略。
[CU011, CU012, CU013, CU014, CU032, CU038]6.4 怀疑视角:工业买家仍会惩罚薄弱集成和站不住的 ROI 账
Rhoda 的公开客户故事应放在工业采购的怀疑视角下看。NIST 指出,潜在制造业用户中只有少数采用了机器人,因为买家在混乱的车间现场里仍难确保集成、性能和互操作性。独立行业资料还指出,如果每个地点都需要定制代码,跨多设施扩展机器人部署会变得贵到难以承受;仓库自动化项目也常常失败,因为 ROI 论证低估了 WMS 集成、停机、培训和变更管理成本。Monocle 2026 年的反向观点在这里尤其相关:许多仓库自动化项目失败,并不是因为机器人技术上做不到,而是因为商业案例建立在错误假设上,机器人周围的运营栈也没有准备好。这个提醒对 Rhoda 很重要,因为公司明确瞄准可变、异常密集的工作流,集成细节昂贵,边缘案例也会快速增多。相邻仓库自动化市场的可比运营商已经发布具名证据点——Amazon 授权 Covariant 技术、KNAPP 公开延长与 Covariant 的合作关系、GXO 公开试点 Dexterity 系统——这意味着该品类中,公开的可背书客户验证门槛实质上高于 Rhoda 当前披露。实际含义很直接。Rhoda 的演示足以支撑继续尽调,但可投资客户证据的门槛应是具名背书、可量化正常运行时间、干预率和站点级 ROI,而不只是有说服力的视频和有利于合作伙伴的叙事。[CU026, CU027, CU028, CU029, CU030, CU031]
| 缺口 | 当前公开状态 | 为何阻碍形成确信 | 补齐条件 |
|---|---|---|---|
| 具名推荐客户 | 官方未披露具名账户 | 无法验证采购行为或用户满意度 | 两个可现场访谈的客户推荐人,包含职务和部署细节 |
| 站点级 ROI 和回本周期 | 未公开 ROI 指标 | 无法区分商业规模与技术上有趣的演示 | 包含人力、正常运行时间和回本数据的客户评分卡 |
| 从试点到生产的转化 | 只有扩张表述 | 没有转化数据,管线质量无法检验 | 试点、付费部署和扩张的分阶段漏斗 |
| 客户集中度 | 未披露客户数或收入结构 | 单一账户敞口可能很大,但外部看不见 | 前五大账户敞口和垂直行业结构 |
每一行都是投资者尽调阻断项或重大不确定性,源自 Rhoda 强工作流证据与薄账户披露之间的不匹配。
[CU024, CU025, CU037, CU038]6.5 图表
07风险
7.1 风险格局排序
Rhoda 最严重的风险都来自一个简单观察:公司正试图在混乱的工业场景中商业化通用机器人智能,但公开记录还没有显示成熟装机基础。这形成了多层风险栈。第一层是技术稳健性:DVA 或许比许多替代方案更省数据、更擅长长上下文推理,但真实工厂或仓库会惩罚每一次失败,方式是基准测试永远模拟不出来的。第二层是安全和责任。机器人一旦在人周围做出高速闭环物理决策,监督不足或故障监测薄弱就可能带来工伤、财产损失和产品责任敞口。第三层是商业化执行:Rhoda 的公开客户信息仍未具名,投资者还无法验证转化、集中度或续约。第四层是依赖风险:硬件无关模型扩大 TAM,但也把成功押在外部硬件、集成和客户运营栈上。最后是市场和估值风险。具身 AI 和人形机器人热度正在上升,但赛道拥挤、昂贵,规模化收入证据仍薄。图 FR001 对当前公开风险栈排序。[CR001, CR002, CR003, CR010, CR026, CR028]
按发生概率、影响、缓释成熟度和剩余严重性,排出 Rhoda 最重要的公开风险。
该矩阵基于公开证据和行业背景给出分析性排序;不能替代内部事故、利润率或客户漏斗数据。
[CR010, CR018, CR022, CR026, CR034, CR041]7.2 安全、监管和责任风险
即便 Rhoda 仍未披露许多运营细节,法律和安全负担也是真实的。OSHA 指出,机器人事故经常发生在设置、维护、测试和调整等非常规状态;OSHA 还指出,美国仍没有专门针对机器人的 OSHA 标准。这意味着部署方必须从更宽泛的机器防护和工作场所安全义务中拼出安全体系,而不能依靠一套单一清晰的规则手册。NIOSH 的机器人中心之所以存在,正是因为协作式、移动式和 AI 驱动机器人的伤害监测与安全实践仍在演进。欧洲又增加了一层:EU AI Act 明确把 AI 监管与健康和安全危害相连,新的 Machinery Regulation 也纳入了 AI 驱动安全功能和网络安全。现有 ISO 标准已经勾勒出安全设计、系统集成和协作机器人预期。法律上最重要的一点是,Rhoda 不能假设一旦出事,下行风险都由客户吸收。Brookings 解释说,过失、设计缺陷和未尽警示义务都可能适用于 AI 系统;Harvard 的监督分析也认为,除非真正具备稳健性、监控和协作系统,否则人在回路并不是充分抗辩。由于 Rhoda 没有公开披露认证、监控框架或保险条款,剩余责任敞口仍高。[CR011, CR012, CR013, CR014, CR015, CR016]
| 风险 | 司法辖区 | 状态 | 可能性 | 严重性 | 缓解成熟度 | 剩余敞口 | 尽调路径 |
|---|---|---|---|---|---|---|---|
| 工人受伤 / OSHA 执法 | 美国 | 当前相关 | 中 | 高 | 公开披露低 | 高 | 获取现场安全规程、防护设计和事故日志 |
| EU AI Act 下的 AI 系统义务 | 欧盟 | 框架已生效 | 中 | 中高 | Unknown | 中高 | 将计划中的欧盟用途映射到高风险和部署方义务 |
| 机械合规与 AI 驱动安全功能 | 欧盟 | 2027 年起强制 | 中 | 高 | Unknown | 高 | 要求提供合规评估计划和公告机构策略 |
| 产品责任 / 未尽警示义务 | 美国和欧盟 | 始终相关 | 中 | 高 | Unknown | 高 | 审阅保险、赔偿、警示和责任分配 |
各行按对投资者的当前实际严重性排序。公开证据确认法律框架存在,但不能确认 Rhoda 当前合规成熟度。
[CR011, CR012, CR013, CR014, CR015, CR016]7.3 技术稳健性、数据飞轮和集成风险
Rhoda 的核心论点在逻辑上自洽:用互联网规模视频建立丰富的运动先验,把预测转化为闭环动作,再让真实部署形成复利式数据飞轮。但从论点到运营公司,中间仍有缺失的桥板。公司称可以用大约 10 到 20 小时机器人数据学会任务,也展示了几个有说服力的概念验证;但 MTBF、干预率、安全认证和生产正常运行时间的公开证据缺席。借助视频生成提升可解释性,可能帮助工程师调试策略行为,但不能证明模型的可审计性足以用于受监管或贴近工人的环境。商业化模式也带来集成风险。Reuters 和 RoboHorizon 都把 FutureVision 描述为可跨大量机器人硬件或现有机队运行;这在战略上有吸引力,但也让 Rhoda 依赖 OEM、集成商、客户 WMS 和 MES 环境,以及公司没有公开列举的站点特定异常处理。NIST 和行业互操作性评论都说明,工厂常常不是因为机器人想法错误而失败,而是集成、标准和异常处理先断裂。如果 Rhoda 不能把未具名演示转化为可重复、可引用的部署,数据飞轮论点会很快变弱。[CR004, CR005, CR006, CR007, CR008, CR009]
| 故障模式 | 可能性 | 严重性 | 缓解成熟度 | 剩余敞口 | 未解决缺口 |
|---|---|---|---|---|---|
| 闭环策略在生产中的长尾物理边界案例上失效 | 中 | 高 | 早期 | 高 | 没有公开干预率或事故记录 |
| 可解释性工具不足以支撑安全验证 | 中 | 中-高 | 早期 | 中-高 | 没有公开认证或监测框架 |
| 客户现场集成瓶颈限制实际吞吐 | 高 | 高 | 低-中 | 高 | 具名集成商和现场架构未披露 |
| 报告、变更管理或异常处理侵蚀 ROI | 高 | 中-高 | 低 | 高 | 没有公开上线后的运营指标 |
运营风险按公开记录已证明和未证明的事项来界定。这里的高可能性指行业常见失效模式, 并不代表 Rhoda 已确认失败。
[CR007, CR008, CR009, CR010, CR020, CR021]| 依赖项 | 交易对方 / 类别 | 作用 | 集中度 | 失效情景 | 严重性 | 缓释措施 | 剩余敞口 |
|---|---|---|---|---|---|---|---|
| 第三方机器人硬件 | 未披露 OEM / 客户设备群 | 物理执行层 | 未知但结构上重要 | 软件表现良好,但本体、可靠性或安全栈在现场失效 | 高 | 保持硬件无关的选择权 | 高 |
| 系统集成商与客户 OT/IT 栈 | 集成商、WMS、MES、设施控制 | 现场部署与异常处理 | 高 | 试点跑通一次,但无法以经济方式跨设施扩展 | 高 | 企业集成手册 | 高 |
| 真实场景部署伙伴和客户 | 未具名工业客户 | 数据飞轮来源 | Unknown | 缺少可背书部署,削弱护城河和模型改进闭环 | 高 | 更多试点和部署 | 高 |
| 监管机构 / 标准生态 | OSHA、NIOSH、EU、ISO 机构 | 合规背书与安全基线 | 中 | 规则收紧快于商业化准备 | 中-高 | 合规规划和标准对齐 | 中-高 |
集中度往往未知,因为 Rhoda 公开材料没有披露底层部署栈或客户名单。未知不等于低。
[CR025, CR026, CR027, CR028, CR037, CR040]展示技术和安全失败如何传导到客户信任、融资压力和估值下行。
DAG 将复杂商业化过程简化为尽调中最重要的因果链。
[CR009, CR010, CR018, CR019, CR026, CR041]梳理 Rhoda 必须依赖哪些外部系统,才能把有前景的智能层变成规模化商业部署。
依赖图强调,即便 Rhoda 采取偏软件的策略,仍要依赖实体、流程和监管系统;这些系统位于 Rhoda 直接公开披露边界之外。
[CR027, CR028, CR037, CR040]7.4 竞争、估值、不透明度和否决标准
即便 Rhoda 技术上跑通,公司仍要在一个可能惩罚延迟和含混的市场中取胜。具身 AI 和人形机器人融资已经异常激进。CNBC 援引 Barclays,把 2035 年的人形机器人市场描绘为潜在巨大,但当前基数仍小,中国已经主导制造规模和成本结构。Humanoids Daily 更广泛的竞争报道把 Rhoda 放在一个拥挤队列中,与 Figure、Tesla、1X 和其他资金充足的入局者并列;即便态度正面的行业观察者也警告,商业化路径漫长,工程陷阱很多。Rhoda 还承受披露不透明的折价。官方材料始终把 2026 年 3 月轮次称为 Series A,但至少一家二级媒体称其为 Series B;关于 Rhoda 主要是硬件无关的大脑层,还是也会自研硬件,公开叙事同样混杂。这些不一致并不致命,但正是会扭曲估值可比、增加尽调难度的含混。实际否决标准因此必须可量化:无法拿出具名可背书客户、缺少可见安全或合规里程碑、试点转付费部署乏力,或证据显示更广泛市场商品化速度快于 Rhoda 构建自有部署数据护城河的速度。[CR029, CR030, CR031, CR032, CR034, CR035]
| 角色 / 职能 | 依赖或缺口 | 可能性 | 严重性 | 缓释措施 | 尽调路径 |
|---|---|---|---|---|---|
| 创始与战略领导层 | Jagdeep Singh 仍是公开层面的主要运营代表 | 中 | 高 | 有更宽的高管梯队 | 索取继任和授权分工图 |
| 安全 / 合规负责人 | 未公开具名安全或合规负责人 | 中 | 高 | 可能在内部已有 | 索取安全、法务和现场运营组织架构图 |
| 现场部署组织 | 未公开披露现场运营或可靠性职能 | 中 | 高 | 可能嵌在工程团队中 | 索取部署团队人数和职责 |
| 商业执行 | 客户推进动作可见,但客户证据仍未具名 | 高 | 高 | 新资金支持招聘和试点 | 索取漏斗、转化和可背书数据 |
问题不是 Rhoda 缺人才,而是公开披露明显偏向研究领导层,部署、合规和商业运营能见度不足。
[CR002, CR003, CR029, CR039]| 风险 | 可监测触发因素 | 阈值 / 事件 | 行动含义 |
|---|---|---|---|
| 安全 / 责任 | 具名认证或事故披露 | 没有可见认证进展,或发生任何严重事故 | 升级尽调,并下调采用率假设 |
| 商业证明 | 具名背书客户和转化数据 | 下次刷新时仍没有具名背书客户或试点转化证据 | 将客户护城河逻辑视为未证实 |
| 集成风险 | 可复制的多站点部署 | 仍只依赖一次性评估 | 下调规模化倍数和部署节奏假设 |
| 竞争 / 估值风险 | 行业融资和部署速度 | 同业拿到大型具名部署,而 Rhoda 仍不透明 | 假设相对物理 AI 同业折价 |
| 披露不透明 | 融资轮次、战略和合规叙事的一致性 | 外部描述持续冲突,管理层没有澄清 | 提高治理和执行折价 |
触发因素设计成刷新时可从外部监测。它们不是失败预测,而是需要重做投资假设的检查点。
[CR026, CR031, CR032, CR038, CR039, CR042]7.5 图表
08估值
8.1 当前轮次背景:巨额 Series A 轮,但基本面图景仍不透明
Rhoda 的 2026 年 3 月轮次首先因规模而突出。$450 million 的 Series A 轮是机器人和具身 AI 品类中最大的早期融资之一,多家二级媒体还把估值放在约 $1.7 billion。这个绝对价格并不是本周期最极端的,但对于一家没有公开披露收入分母、毛利率画像、客户数或合同经济性的公司,仍是很高的价格。换句话说,这一轮足够大,能显示投资人信念很强;但透明度不够,外部投资人无法把估值转化为严谨倍数。 围绕这一轮的公开证据也有质量问题。Rhoda 和 Wilson Sonsini 称该融资为 Series A,而至少一家二级媒体把它标成 Series B。这种不一致不改变募资金额,但有助于提醒我们:具身 AI 私募市场元数据噪音很大,尽调应优先采用官方来源,而不是聚合器叙事。同样的模式也体现在缺失项上:Rhoda 披露工业部署和客户试点,但没有任何公开来源给出付费客户名称、ARR、定价结构或留存行为。 积极解读是,市场在为一个可信团队攻坚难题的期权价值定价。Rhoda 领导层阵容很强,其声称的软件层野心在战略上也有吸引力。消极解读是,当前价格主要靠叙事和未来可选性支撑,而不是靠已观察到的商业化支撑。对于正在决定是否追高进场的投资者,这一区别比本轮标题金额本身更重要。[CV001, CV002, CV003, CV004, CV005, CV006]
| 指标 | 当前判断 | 决策含义 |
|---|---|---|
| 建议 | 观察 / 继续研究 | 没有新证据,不追明显更高的入场价 |
| 信心 | 中低 | 品类顺风真实存在,但公司层面的经济性仍不透明 |
| 风险评级 | 高 | 未披露收入和客户集中度,让投资测算基础脆弱 |
| 估值立场 | 偏满但不是行业最高 | 低于超级融资轮同业,但相对已披露基本面仍然偏贵 |
建议明确对价格和证据敏感。它不是对团队或品类的泛化质量评分。
[CV002, CV005, CV020, CV027, CV039, CV040]| 论点 | 当前证据 | 什么会改变判断 |
|---|---|---|
| 投资逻辑:品类顺风大 | 2026 年 Q1 物理 AI 融资和 2026 年 AI 预算扩张显示投资人胃口仍有韧性 | 如果出现具名企业部署和复购客户,判断会进一步增强 |
| 投资逻辑:软件层定位有战略价值 | Rhoda 将 FutureVision 定位为可授权、跨硬件的智能层 | 如果披露定价和附着率,判断会增强 |
| 投资逻辑:团队强,方向贴近工业场景 | 领导层履历和生产环境声明方向上利好 | 如果有客户背书和正常运行时间数据,判断会增强 |
| 反向逻辑:没有公开收入分母 | 保留的公开来源没有收入、利润率或定价披露 | 如果 Rhoda 披露 ARR、利润率和续约行为,反向逻辑会削弱 |
| 反向逻辑:私募市场噪音高 | 二手来源对融资轮次信息说法不一,许多同业按不透明私募估值交易 | 如果第三方数据趋同,私募估值也能映射到公开市场可比公司,反向逻辑会削弱 |
| 反向逻辑:商业化可能落后于叙事 | 同业报道显示,一些头部公司仍缺商业化时间表 | 如果 Rhoda 发布可复制部署经济性,反向逻辑会削弱 |
反向逻辑不是假设;它基于当前披露缺口和可比公司报道。
[CV003, CV004, CV005, CV006, CV007, CV012]从品类顺风和证明缺口,推导到观察 / 继续研究立场的决策链。
该流程代表分析师基于保留公开证据形成的判断,不是公司发布的决策框架。
[CV002, CV005, CV007, CV020, CV027, CV028]8.2 未上市具身 AI 可比公司:Rhoda 低于巨额轮次,但板块明显虚高
Rhoda 应放在当前具身 AI 估值阶梯中估值,而不应只对标成熟工业自动化企业。在这条阶梯上,Figure 是明确上限,2025 年估值约 $39 billion。Skild 从 2024 年的 $1.5 billion 升至 2026 年超过 $14 billion,同时披露收入约 $30 million;这个倍数只有在投资者相信它能成为默认机器人“大脑”平台时才说得通。据报道,Physical Intelligence 在 2026 年寻求超过 $11 billion 的估值,而几个月前估值约 $5.6 billion;与此同时,TechCrunch 报道其没有商业化时间表。Apptronik 达到 $5 billion,Dexterity 达到 $1.65 billion。FieldAI 融了大量资金,甚至没有公开披露估值。 与这组相比,Rhoda 的 $1.7 billion 绝对值看起来不高。这是乐观论点:投资者用远低于 Figure、Skild、Physical Intelligence 或 Apptronik 的价格,买到同一大类敞口——面向混乱真实世界任务的通用机器人智能。不过,反向逻辑是,那些更高估值未必证明 Rhoda 便宜;它们也可能只是揭示整个同业组已经被拉得过高。 这正是怀疑来源重要的地方。TechCrunch 关于 Physical Intelligence 的报道指出,其没有商业化时间表。Eilla 的机器人估值手册认为,更接地气的仓储和内部物流机器人企业通常落在低个位数到中个位数 EV/revenue 区间,服务占比高的集成商甚至更低。对于一家收入前的前沿模型公司,这不是完美的逐项可比基准,但它是有用的纪律锚点。当前具身 AI 市场显然奖励叙事、人才和战略可选性。它还没有证明,所有这些私募估值都能转化为持久的公开市场价值。[CV007, CV008, CV009, CV010, CV011, CV012]
| 情景 | 核心假设 | 估值 / 回报逻辑 | 关键风险 | 概率信号 |
|---|---|---|---|---|
| 乐观 | Rhoda 将试点转化为可见授权收入,证明可复制性,并成为可信的多硬件软件平台 | 24 个月内 $3B-$5B;当前估值成为可接受的早期入场点 | 执行复杂度、竞争、安全、客户集中度 | 低至中 |
| 基准 | Rhoda 披露客户和收入,但仍处商业化早期,站点扩展不均衡 | $1.5B-$2.2B;当前估值大致合理,有温和上行 | 销售周期长、利润率不确定、预算闸门 | 中 |
| 悲观 | 收入仍未披露,行业倍数压缩,证据更强的同业重设预期 | $0.9B-$1.3B;较当前估值有下行 | 私募市场重估、试点转化弱、采购放慢 | 中 |
| 价格纪律 | 投资人拒绝在新披露前加价 | 回报来自等待分母清晰,而不是追即时动能 | 如果 Rhoda 快速执行,可能错过上行 | 审慎默认 |
情景区间刻意粗略,因为公开证据不足以支持对 Rhoda 做精细 DCF 或 ARR 建模。
[CV035, CV036, CV037, CV038, CV039, CV040]| 可比公司 | 指标 / 状态 | 观察到的估值或倍数 | 可比性 | 局限 |
|---|---|---|---|---|
| Rhoda AI | 私募轮次,收入未披露 | 据报道估值 $1.7B | 标的公司;物理 AI 领域大型 Series A 轮 | 无公开收入分母 |
| Dexterity | 私募轮次(2025 年 3 月) | 投后估值 $1.65B | 工业机器人智能领域最接近、且已披露的低估值同业 | 任务特定模型,不是同一种平台叙事 |
| Apptronik | 私募轮次(2026 年 2 月) | 估值 $5B | 融资热度强的人形机器人可比公司 | 偏硬件的人形机器人模式不同于 Rhoda 的软件层定位 |
| Physical Intelligence | 私募轮次 / 谈判(2025-2026) | $5.6B 至 >$11B | 纯物理 AI / 机器人“大脑”叙事可比公司 | 商业化时间表仍不清楚 |
| Skild AI | 私募轮次(2026 年 1 月) | > $14B;约 467x 过去 12 个月收入 | 已披露机器人“大脑”估值倍数中最高 | 规模高得多,且披露了未经审计收入 |
| Figure AI | 私募轮次(2025 年 9 月) | $39B | 物理 AI 热度和部署证明的上界 | 人形机器人全栈硬件 / 软件经济性 |
| Symbotic | 上市仓储自动化可比公司 | 2.27x 销售额;$6.03B 市值 | 最适合做仓储自动化合理性校验的倍数 | 成熟度高得多,且积压订单更重 |
| Zebra Technologies | 上市自动化 / 数据采集可比公司 | 2.10x 销售额;$11.06B 市值 | 有用的工业和物流软件 / 硬件基准 | 企业产品组合更宽 |
| Rockwell Automation | 上市工业自动化可比公司 | 5.72x 销售额;$49.71B 市值 | 成熟工业自动化中的公开市场上沿倍数 | 成熟且盈利能力强得多 |
私募估值和公开市场倍数不能直接比较。此表用于三角校验和价格纪律,不是做虚假精确的平均。
[CV002, CV008, CV010, CV011, CV013, CV014]8.3 上市自动化公司可比与情景测算:当前估值靠对未来收入的信念,不靠当下收入证据
上市自动化企业并不是 Rhoda 产品模式的直接可比,但它们是估值纪律能找到的最好护栏。截至报告运行日,Yahoo Finance 显示 Symbotic 约为 2.27x 销售额,Zebra 约为 2.10x 销售额,Rockwell 约为 5.72x 销售额。这些公司成熟得多,有真实收入分母和已建立的商业历史,因此不应被用来直接给 Rhoda “估值”。但它们能作为理性校验:当公开市场投资者真正看得见收入时,规模化自动化企业是什么样。 含义很直接。如果 Rhoda 要用 5x 收入支撑 $1.7 billion——高于 Symbotic 和 Zebra、低于 Rockwell 的溢价——就需要大约 $340 million 收入。即便按 10x 收入,也仍需要约 $170 million。公开来源并没有显示 Rhoda 今天接近这个水平,因为公开来源根本没有给出任何 Rhoda 收入数字。这并不意味着 Rhoda 最终无法长进这个估值。它意味着当前估值并没有扎在一个可观察分母上。 由此可以形成三情景框架。悲观情景假设私募估值回归常态,Rhoda 也无法把试点转化为已披露的经常性收入;公允价值随后压缩到大约 $0.9 billion 到 $1.3 billion。基准情景假设 Rhoda 把技术承诺转化为更清晰的付费部署和收入披露,支撑当前估值附近或略高的水平。乐观情景假设 Rhoda 成为可信的多硬件软件平台,并出现可见授权增长,未来 24 个月把价值抬升至 $3 billion 到 $5 billion 区间。三种情景都必然粗略,因为缺失分母才是核心问题。[CV021, CV022, CV023, CV024, CV027, CV028]
| 触发因素 | 阈值 | 对投资逻辑的传导 | 行动含义 |
|---|---|---|---|
| 到下一轮融资周期仍无具名付费部署 | 仍没有客户名称、定价或 ARR 披露 | 软件经济性逻辑坍缩为纯叙事风险 | 不加价;先要求硬客户证据 |
| 证据更强的同业发生行业重估 | 更强同业出现平轮 / 降估值轮或重大估值下调 | 压缩 Rhoda 可享受的叙事溢价 | 将公允价值重估至悲观情景 |
| 试点经济性无法规模化 | 一旦披露后显示实施成本高或毛利率弱 | 削弱平台软件逻辑 | 改按服务占比高的集成商经济性看待 |
| 模型在真实站点的安全 / 正常运行时间上失效 | 独立 KPI 数据与演示说法相矛盾 | 损害企业采购概率 | 暂停尽调,直到可靠性证据出现 |
| 股权结构或投资人保护不透明 | 出现意外优先股堆叠或未来大幅稀释 | 即使公司运营执行到位,也会压低上行 | 重算预期回报,或直接退出 |
这些是在当前报告估值及以上的叫停触发因素。如果入场价低得多,一部分会变成监测项,而不是硬性停止条件。
[CV012, CV025, CV026, CV041, CV042]在不同收入倍数假设下,支撑 $1.7 billion 估值所需的年收入。
数值为所需年收入,单位为 USD millions。该图展示的是缺失分母问题,而不是声称 Rhoda 能或不能达到这些水平。
[CV022, CV023, CV024, CV025, CV026, CV038]仅用公开证据和可比公司纪律,为 Rhoda 划出悲观、基准、乐观估值区间。
区间刻意保持粗略,因为 Rhoda 不披露收入、定价或利润率分母。当前据报估值仅供参考。
[CV002, CV035, CV036, CV037, CV039]8.4 建议、投资逻辑破坏点和最终尽调问题
以公开证据得出的最干净结论是观察 / 继续研究,而不是激进追高。品类顺风很明显:劳动力短缺持续存在,自动化预算仍有意义,AI 预算在扩张,大型投资者也显然愿意为具身 AI 平台赌注融资。Rhoda 还受益于强领导层信号,以及具有战略吸引力的“机器人智能层”叙事。如果公司后续证明其授权经济性在制造和物流中具备持久性,今天的价格可能会像一个可接受的早期路标,而不是峰值。 但当前证据还不足以用高置信度支持这一结果。决定性缺失项都很基础:具名付费客户、定价结构、经常性收入、毛利率、部署留存,以及模型能在精挑细选的评估之外稳定工作的证明。在披露这些之前,投资案例对价格和证据敏感,而不是只对品类敏感。换句话说,相信具身 AI 还不够;问题是 Rhoda 自己能否把这种信念转化为经得起公开市场审视的软件经济性。 如果下一轮融资周期仍缺少具名付费部署和收入披露,或者证据更扎实的同业出现平轮或下调轮、重新锚定整个板块,投资逻辑会最快破裂。因此,尽调议程很直接。在显著高于据报当前估值付款之前,投资者应要求看到站点级付费使用证据、软件抽成率清晰度、实施成本和毛利数据,以及能说明正常运行时间和扩张行为的客户背书。没有这些,审慎立场就是保持跟踪、学习,等待证据,而不是追逐兴奋。[CV005, CV027, CV028, CV029, CV030, CV031]
| 主题 | 缺失证据 | 重要性 | 负责人 / 尽调路径 |
|---|---|---|---|
| 收入模型 | 定价依据、ARR/MRR 和客户数 | 需要它才能把估值转化为真实倍数 | 与管理层开展财务 + GTM 尽调 |
| 客户证明 | 具名付费客户、站点数量和扩张数据 | 区分试点和可复制商业需求 | 背书访谈和队列分析 |
| 单位经济性 | 毛利率、实施成本、支持负担、算力成本 | 决定 Rhoda 更像软件公司,还是服务占比高 | 管理层数据室 + 部署模型审查 |
| 可靠性 | 第三方正常运行时间、安全和错误率记分卡 | 企业买家重视可靠性胜过新颖性 | 客户 KPI 包 / 技术尽调 |
| 股权结构 | 优先权、清算堆叠和稀释条款 | 即使名义估值可接受,也会影响实际回报 | 法务 / 融资尽调 |
| GTM 路径 | 集成商、OEM 和直销渠道组合 | 厘清采用速度和利润率获取 | 渠道访谈 + 合同审查 |
前三个问题中任意一个都可能实质改变当前估值下的建议,因为 Rhoda 的公开证据集异常缺分母。
[CV005, CV027, CV028, CV040, CV041, CV045]按 IC 口径,对 Rhoda 在当前据报估值下的市场、证明、经济性和估值支撑打分。
分数是分析师按 1-5 分制给出的判断,仅使用截至运行日期保留的公开证据。
[CV005, CV007, CV020, CV027, CV028, CV039]8.5 图表
免责声明
本报告基于截至 2026-06-09 的公开信息,是分析性尽职调查材料,不构成投资建议。
证据索引
| 编号 | 陈述 | 可信度 | 来源 |
|---|---|---|---|
| CO001 | Rhoda AI publicly launched on 2026-03-10 after 18 months in stealth. | 高 | SO004, SO009, SO014 |
| CO002 | Official launch materials identify Rhoda as Palo Alto, California-based. | 高 | SO004, SO015, SO019 |
| CO003 | Rhoda positions itself as a builder of general-purpose robot foundation models for commercial and industrial environments. | 高 | SO001, SO004 |
| CO004 | FutureVision is the company’s intelligence layer for robotic systems and is expected over time to be licensed across partner hardware and software platforms. | 高 | SO001, SO004, SO013 |
| CO005 | Rhoda’s Direct Video Action architecture pre-trains on internet-scale video and then maps predicted future video into robot actions in a closed loop. | 高 | SO002, SO004, SO013 |
| CO006 | The official site says Rhoda first pre-trains on over a million videos and then post-trains on 1–10 hours of trajectory data. | 高 | SO001, SO002 |
| CO007 | Rhoda’s research note says the DVA approach can learn complex long-horizon tasks with roughly 10 hours of robot data. | 高 | SO002, SO004, SO013 |
| CO008 | Rhoda publicly showcases returns processing, bearing decanting, container breakdown, and human-demo following as representative workflows. | 高 | SO001, SO002 |
| CO009 | The company says it works with customers across automotive, manufacturing, logistics, and ecommerce. | 高 | SO001, SO004 |
| CO010 | Rhoda says a recent high-volume manufacturing evaluation completed a component-processing workflow in under two minutes per cycle without human intervention. | 高 | SO004, SO014, SO018 |
| CO011 | Rhoda presents itself as a hybrid model company because it markets FutureVision as licensable software while also operating its own robotic systems. | 中 | SO001, SO004, SO019 |
| CO012 | The home page also describes a Rhoda robot platform with custom actuators, safety-rated vision, a 25kg rated payload, and 40kg peak payload. | 中 | SO001 |
| CO013 | Rhoda’s team page names Jagdeep Singh as CEO and co-founder. | 高 | SO005, SO004 |
| CO014 | Rhoda’s team page names Eric Chan as Chief Scientist and Gordon Wetzstein as Scientific Advisor. | 高 | SO005, SO004 |
| CO015 | Public launch coverage says Eric Ryan Chan previously served as a generative model architect at WorldLabs and is a Stanford researcher. | 高 | SO004, SO011, SO014 |
| CO016 | Gordon Wetzstein’s Stanford page says he has been a Rhoda AI co-founder since October 2024. | 中 | SO020 |
| CO017 | Wetzstein’s Stanford biography describes him as a Stanford EE associate professor and director of the Stanford Physical and Spatial Intelligence Lab. | 中 | SO020 |
| CO018 | Rhoda’s team page also publicly names Andrew Wooten, Changan Chen, Steve Tirado, and Alex Bergman among the leadership team. | 中 | SO005 |
| CO019 | Rhoda’s public Ashby board listed 33 open positions in Palo Alto across research, software, hardware, business, and operations when fetched for this run. | 中 | SO008 |
| CO020 | The public team page enumerates 62 named team members, which is a lower-bound people signal rather than a full employee census. | 中 | SO005 |
| CO021 | Rhoda AI Corporation appears in California registry data as an active Delaware stock corporation incorporated on 2024-08-01. | 中 | SO021 |
| CO022 | The registry page lists a San Jose registered address, which differs from Palo Alto operating-location language in launch materials. | 中 | SO021, SO004 |
| CO023 | Rhoda’s public news page currently shows only a single press-release entry dated 2026-03-10. | 中 | SO003 |
| CO024 | Rhoda announced a $450 million Series A financing on 2026-03-10. | 高 | SO004, SO009, SO012 |
| CO025 | Multiple secondary sources value the March 2026 round at about $1.7 billion. | 中 | SO010, SO016, SO017, SO019, SO024, SO026 |
| CO026 | Official and syndication sources publicly name Capricorn, Khosla, Leitmotif, Matter Venture Partners, Mayfield, Premji Invest, Prelude Ventures, Temasek, Xora, and John Doerr among Rhoda’s backers. | 高 | SO004, SO009, SO011, SO013 |
| CO027 | Public sources disagree on lead attribution for the round, while Rhoda’s own release lists backers but does not name a single lead investor. | 中 | SO004, SO012, SO019, SO024 |
| CO028 | Wilson Sonsini describes the financing as a Series A round led by a multi-name syndicate rather than a single investor. | 中 | SO012 |
| CO029 | Several secondary write-ups describe Premji Invest as the lead investor in the March 2026 round. | 中 | SO017, SO019, SO024 |
| CO030 | Some secondary trackers misclassify the March 2026 financing as Series B even though Rhoda’s own materials call it Series A. | 中 | SO016, SO026 |
| CO031 | Public leadership visibility is concentrated in Jagdeep Singh plus the Eric Chan and Gordon Wetzstein research axis, creating clear key-person concentration risk. | 中 | SO004, SO005, SO020 |
| CO032 | No public board of directors or formal governance structure appears in the retrieved company materials. | 中 | SO003, SO004, SO005, SO007 |
| CO033 | The retrieved public record does not disclose revenue, named customer count, exact headcount, or public pricing. | 中 | SO001, SO003, SO004, SO009 |
| CO034 | The company talks about industrial partners and production environments but does not publicly identify named enterprise customers. | 中 | SO001, SO004, SO013, SO014 |
| CO035 | robotics.press characterizes Rhoda as pre-revenue with no independently validated deployments or disclosed unit economics. | 低 | SO025 |
| CO036 | robotics.press also flags the absence of publicly disclosed operations or field-deployment executives as an execution risk for industrial rollouts. | 低 | SO025 |
| CO037 | TechStackIPO marks Rhoda’s profile as verification pending and introduces incorrect stage history, illustrating noisy third-party metadata around the company. | 低 | SO026 |
| CO038 | Tracxn lists Rhoda as founded in 2024 and at roughly 60 employees as of late March 2026, giving an external but not primary-verified scale signal. | 低 | SO022 |
| CM001 | Because Rhoda describes FutureVision as an intelligence layer that can be licensed across different robotic hardware and software platforms, the company’s monetizable category should be treated as robot-intelligence software rather than total robot hardware spend. | 高 | SM001, SM003, SM004 |
| CM002 | Rhoda describes FutureVision as a robotic intelligence system based on video-predictive control. | 高 | SM001, SM003, SM005 |
| CM003 | Rhoda says FutureVision is an intelligence layer that can be licensed across different robotic hardware and software platforms. | 高 | SM003, SM004, SM005 |
| CM004 | Rhoda states that it works with industrial partners across manufacturing and logistics. | 高 | SM003, SM006 |
| CM005 | Rhoda says its models are pre-trained on internet-scale video and then post-trained on smaller amounts of robot data. | 高 | SM003, SM005 |
| CM006 | Rhoda claims its closed-loop Direct Video Action architecture updates behavior continuously as conditions change. | 中 | SM003, SM005 |
| CM007 | Rhoda says the strong motion prior from video pretraining can reduce new-task data needs to as little as about ten hours of teleoperation. | 中 | SM003, SM005 |
| CM008 | Rhoda says it completed a component-processing workflow in under two minutes per cycle without human intervention in a recent manufacturing evaluation. | 中 | SM003, SM005 |
| CM009 | Rhoda’s public team page names seven senior leaders and more than sixty individual team members. | 中 | SM002 |
| CM010 | The global AI robots market is projected to grow from $6.11 billion in 2025 to $33.39 billion in 2030. | 中 | SM007 |
| CM011 | MarketsandMarkets projects a 40.4% CAGR for the AI robots market from 2025 to 2030. | 中 | SM007 |
| CM012 | The same AI robots market source expects hardware to account for 61% of the market in 2025. | 中 | SM007 |
| CM013 | MarketsandMarkets lists software and services as explicit offerings within the physical AI market taxonomy. | 中 | SM025 |
| CM014 | MarketsandMarkets projects the physical AI market to grow from $1.50 billion in 2026 to $15.24 billion in 2032. | 中 | SM025 |
| CM015 | Mordor Intelligence sizes the warehouse automation market at $34.17 billion in 2026 and $65.74 billion in 2031. | 中 | SM008 |
| CM016 | Modern Materials Handling says organizations invested about $21 billion in warehouse automation in 2023 and expects more than $90 billion by 2033. | 中 | SM017 |
| CM017 | IFR recorded 542,000 industrial robot installations in 2024 worldwide. | 中 | SM009 |
| CM018 | IFR estimated 4.664 million industrial robots were in operational use worldwide in 2024. | 中 | SM009 |
| CM019 | IFR expects global robot installations to rise to about 575,000 units in 2025 and to surpass 700,000 units by 2028. | 中 | SM009 |
| CM020 | Cobots represented 10.5% of industrial robot installations in 2023. | 中 | SM010 |
| CM021 | IFR says cobots are especially attractive for flexible production settings without deep in-house engineering resources. | 中 | SM010 |
| CM022 | McKinsey found that logistics and fulfillment players expect automation to represent 30% or more of capital spending over the next five years. | 中 | SM012 |
| CM023 | McKinsey says capital cost is the top adoption barrier, cited by 71% of industrial respondents. | 中 | SM012 |
| CM024 | McKinsey says 61% of industrial respondents cite lack of automation experience as an adoption barrier. | 中 | SM012 |
| CM025 | McKinsey reports that 62% of industrial respondents prefer vendors that can provide full-service implementation models. | 中 | SM012 |
| CM026 | The U.S. Chamber says durable-goods manufacturing still had about 313,000 unfilled job openings as of April 2025. | 中 | SM014 |
| CM027 | BLS counted 6.95 million hand laborer and material mover jobs in 2024. | 中 | SM015 |
| CM028 | BLS projects about 1.008 million annual openings for hand laborers and material movers over the coming decade. | 中 | SM015 |
| CM029 | BLS says transportation and warehousing accounts for 21% of hand laborer and material mover employment. | 中 | SM015 |
| CM030 | Hy-Tek says warehouse automation is shifting from hardware-driven systems to software-defined environments. | 中 | SM016 |
| CM031 | Hy-Tek says Robotics-as-a-Service is lowering the upfront capital barrier for warehouse automation. | 中 | SM016 |
| CM032 | Hy-Tek describes warehouse execution systems as the central nervous system that synchronizes AS/RS, conveyors, AMRs, and robotics. | 中 | SM016 |
| CM033 | Modern Materials Handling says 92% of buyers rate durability, reliability, and uptime as very important when evaluating automation systems. | 中 | SM017 |
| CM034 | Modern Materials Handling says 95% of buyers view fast service response times as essential in automation selection. | 中 | SM017 |
| CM035 | Modern Materials Handling says average planned spend on materials-handling equipment rises to $1.6 million in 2026 from $1.5 million in 2025. | 中 | SM017 |
| CM036 | StartUs Insights projects the 3PL market to grow from $1.8 trillion in 2026 to $4.3 trillion by 2035 at a 10.1% CAGR. | 中 | SM013 |
| CM037 | StartUs Insights says AI in logistics is growing at 17.44% annually within its tracked 3PL innovation dataset. | 中 | SM013 |
| CM038 | NVIDIA says 90% of surveyed retail and CPG respondents plan to increase AI budgets in 2026. | 中 | SM022 |
| CM039 | NVIDIA says 17% of respondents are already using or evaluating physical AI in retail and supply chain operations. | 中 | SM022 |
| CM040 | UPS says U.S. freight volumes are forecast to grow about 2.3% in 2026. | 中 | SM021 |
| CM041 | UPS says fewer than one third of executives have achieved end-to-end visibility and poor visibility correlates with about 50% higher inventory carrying costs and about 30% longer lead times. | 中 | SM021 |
| CM042 | DHL says 2026 logistics planning is increasingly shaped by autonomous decision-making, sustainability, and elastic logistics rather than by fixed one-time optimization projects. | 中 | SM020 |
| CM043 | Interact Analysis says 2025 uncertainty pushed it to cut its overall warehouse automation forecast, with the mobile robot outlook revised down more sharply than fixed automation. | 中 | SM018, SM019 |
| CM044 | Interact Analysis says warehouse automation revenue in 2024 still grew 1% versus its earlier expectation of a 3% decline. | 中 | SM019 |
| CM045 | Interact Analysis says brownfield retrofits dominate near-term deployments and greenfield projects are more likely to rebound from 2027 onward. | 中 | SM018 |
| CM046 | The Robot Report describes foundation-model robotics as a horizontal strategy that aims to supply a general-purpose brain across robot embodiments and tasks. | 中 | SM023 |
| CM047 | The Robot Report says the next-generation robot AI race is increasingly about data collection and model scale rather than just building hardware. | 中 | SM024 |
| CM048 | A software-centric TAM for Rhoda is materially smaller than the full warehouse automation or industrial robot hardware market because hardware remains the majority share in published AI robot taxonomies. | 中 | SM007, SM015, SM016, SM025 |
| CM049 | Public sources show Rhoda has pilots and production evaluations, but they do not disclose paid customer names, pricing, revenue, or software take rate. | 中 | SM001, SM003, SM004, SM005, SM006 |
| CM050 | The most credible near-term buyers for Rhoda are operations teams and system integrators in manufacturing and logistics that already own hardware budgets but need software to automate higher-variability tasks. | 中 | SM003, SM012, SM016, SM017 |
| CP001 | Rhoda positions FutureVision as an intelligence layer meant to power its own systems and eventually be licensed across multiple robotic hardware and software platforms. | 中 | SP001 |
| CP002 | Rhoda says its data strategy starts with internet-scale video pretraining and then adds smaller amounts of robot data for embodiment-specific post-training. | 高 | SP001, SP002 |
| CP003 | Skild argues a shared model across different robot form factors is necessary because robotics data is too scarce to silo by embodiment. | 中 | SP005 |
| CP004 | Skild announced a $1.4 billion round in January 2026 at a valuation above $14 billion. | 高 | SP004, SP024 |
| CP005 | Skild publicly links its software strategy to real deployments, eight partners, and platform distribution through relationships such as ABB, Universal Robots, Foxconn and NVIDIA. | 中 | SP005, SP017 |
| CP006 | Physical Intelligence describes π0 as a general-purpose robot foundation model that combines images, text and actions to output low-level motor commands. | 高 | SP006, SP008 |
| CP007 | Physical Intelligence says π0 uses internet-scale pretraining and dexterous data collected across eight distinct robots. | 中 | SP006 |
| CP008 | Physical Intelligence reports that π0 outperforms OpenVLA and Octo on its five-task evaluation set. | 高 | SP006, SP008 |
| CP009 | Openpi publishes open-source models, checkpoints and training paths for π0, π0.5 and downstream benchmark variants such as LIBERO and DROID. | 中 | SP007 |
| CP010 | Figure describes Helix as a generalist humanoid vision-language-action system that handles perception, movement and reasoning on board in real time. | 中 | SP009 |
| CP011 | Independent trackers place Figure at roughly a $39 billion valuation with BMW as its flagship industrial deployment proof. | 中 | SP010, SP011 |
| CP012 | TechMarketBriefs frames Figure’s core thesis as vertical integration of robot, AI model and BotQ factory, but highlights a valuation and safety-risk bear case. | 中 | SP011 |
| CP013 | Dexterity says its physical-AI stack is trained on more than 100 million autonomous actions in production. | 中 | SP012 |
| CP014 | Dexterity says its robots already run full shifts at major logistics operators and make millions of autonomous decisions with zero safety incidents. | 中 | SP012 |
| CP015 | FieldAI markets EDGE as one brain across robots, tasks and environments built on a belief world model and risk-aware autonomy. | 中 | SP013 |
| CP016 | FieldAI says it has deployments across three continents and public partnerships with NVIDIA and Boston Dynamics in 2026. | 中 | SP013, SP017 |
| CP017 | Apptronik says Apollo is a general-purpose humanoid aimed first at warehouses and manufacturing plants. | 中 | SP014, SP015 |
| CP018 | Apptronik highlights a robot-as-a-service model and mass-manufacturability rather than a neutral software-licensing strategy. | 中 | SP014, SP015 |
| CP019 | NVIDIA GR00T N1.7 is an open cross-embodiment VLA model that NVIDIA distributes under Apache 2.0. | 中 | SP016 |
| CP020 | GR00T N1.7 uses 20,000 hours of EgoScale human video plus diverse robot data and supports fine-tuning for new embodiments. | 中 | SP016 |
| CP021 | NVIDIA explicitly positions Skild AI and FieldAI as generalized robot-brain developers building on Cosmos world models and Isaac simulation frameworks. | 中 | SP017, SP026 |
| CP022 | Google DeepMind describes Gemini Robotics as a Gemini 2.0-based VLA model focused on generality, interactivity and dexterity. | 中 | SP018 |
| CP023 | Google DeepMind says Gemini Robotics more than doubles performance on its generalization benchmark compared with prior state-of-the-art VLA models. | 中 | SP018 |
| CP024 | Google says Gemini Robotics can adapt across multiple robot types and specifically names Apptronik Apollo as a target embodiment. | 中 | SP018 |
| CP025 | Gemini Robotics-ER is pitched as an embodied-reasoning layer that can connect to existing low-level controllers and is available to testers such as Agility and Boston Dynamics. | 中 | SP018 |
| CP026 | Rhoda differs from VLA-heavy rivals by making causal video prediction and inverse-dynamics translation, rather than language-conditioned action decoding, the center of its control loop. | 中 | SP002, SP018 |
| CP027 | Rhoda’s strongest public technical advantage claim is robot-data efficiency, while Skild, PI, NVIDIA and Google have disclosed more explicit benchmark or platform-comparison artifacts. | 中 | SP002, SP006, SP016, SP018 |
| CP028 | Hardware-agnostic model labs such as Rhoda, Skild, Physical Intelligence and FieldAI pursue broader OEM reach but inherit integration and support burdens across heterogeneous robots. | 中 | SP001, SP003, SP006, SP013 |
| CP029 | Vertically integrated players such as Figure and Apptronik can optimize hardware and control together but are more tied to the economics and pace of one robot family. | 中 | SP009, SP011, SP015 |
| CP030 | Platform incumbents such as NVIDIA and Google can subsidize robotics models with compute, simulation or broader AI-platform revenue, pressuring the pricing power of software-only startups. | 中 | SP017, SP018, SP026 |
| CP031 | Skild’s valuation implies the market rewards neutral robot-brain providers, but its public evidence remains heavier on vision, partner narrative and funding than on peer-reviewed head-to-head benchmarks. | 中 | SP004, SP005, SP025 |
| CP032 | Figure has the strongest disclosed industrial pilot among Rhoda’s adjacent competitors because it has published BMW production metrics rather than only lab or conference demos. | 中 | SP010, SP011 |
| CP033 | Physical Intelligence’s openpi release lowers barriers for developer adoption and experimentation, which is a different moat strategy from Rhoda’s proprietary FutureVision stack. | 中 | SP007, SP001 |
| CP034 | Dexterity’s narrower warehouse/logistics focus gives it deeper production evidence in that niche than broader foundation-model labs currently disclose. | 中 | SP012 |
| CP035 | FieldAI’s public positioning emphasizes inspection, construction and industrial field autonomy, making it a closer rival on industrial reliability than on warehouse manipulation benchmarks. | 中 | SP013 |
| CP036 | Covariant remains an adjacent warehouse-AI reference point, but the official source surface retrieved this run is sparse relative to the newer physical-AI model labs. | 低 | SP027, SP021 |
| CP037 | Public pricing across Rhoda, Skild, PI, NVIDIA, Dexterity and Figure remains mostly opaque, so competitive analysis has to compare deployment models and channel leverage rather than list price. | 中 | SP001, SP003, SP009, SP012, SP016 |
| CP038 | The field’s data strategies now split into internet video plus lightweight robot post-training (Rhoda), simulation plus deployment data (Skild and FieldAI), multi-robot dexterous datasets plus VLM pretraining (PI), production-action logs (Dexterity), and platform-scale human plus robot data (GR00T and Gemini). | 中 | SP002, SP005, SP006, SP012, SP013, SP016, SP018 |
| CP039 | Google and NVIDIA are the most durable competitive threats because they can combine model advances with ecosystem control over simulators, compute or foundation AI. | 中 | SP017, SP018, SP026 |
| CP040 | Rhoda’s differentiation durability depends less on architecture alone than on whether video-first training turns into repeatable real-world deployment data before larger rivals close the robustness gap. | 中 | SP001, SP002, SP005, SP018 |
| CI001 | Rhoda presents FutureVision as an intelligence layer that can be licensed across partner hardware and software platforms over time. | 高 | SI001, SI004 |
| CI002 | The home page simultaneously markets a Rhoda robot platform with custom actuators and safety-rated vision, implying the company is not a pure software wrapper. | 中 | SI001 |
| CI003 | Rhoda’s public use cases cluster around industrial returns processing, automotive decanting, and heavy-container breakdown rather than general consumer robotics. | 高 | SI001, SI004 |
| CI004 | Official materials say Rhoda works with customers across automotive, manufacturing, logistics, and ecommerce. | 高 | SI001, SI004 |
| CI005 | Rhoda says the $450M financing will fund research and engineering, industrial deployments, customer pilots, and team growth. | 高 | SI004, SI007, SI009 |
| CI006 | No retrieved public source discloses revenue, ARR, GMV, or audited financial statements for Rhoda. | 中 | SI001, SI003, SI004, SI007, SI021 |
| CI007 | No retrieved public source discloses pricing, ACV, or standardized contract structure for FutureVision. | 中 | SI001, SI004, SI014, SI021 |
| CI008 | No retrieved public source names enterprise customers or provides a customer count. | 中 | SI001, SI004, SI021 |
| CI009 | Rhoda’s public product positioning implies a hybrid monetization model of software licensing, deployment services, and possibly internally developed systems rather than pure SaaS. | 中 | SI001, SI004, SI019 |
| CI010 | Rhoda’s public Ashby board listed 33 openings, heavily weighted to research and software, which points to a large fixed-cost research and infrastructure base. | 中 | SI020 |
| CI011 | Specific live postings for VP of Hardware, Supply Chain & Logistics Lead, and Inference Infrastructure Engineer show Rhoda is staffing hardware leadership, operations, and compute infrastructure in parallel. | 中 | SI025, SI026, SI027 |
| CI012 | The DVA research note emphasizes web-scale video pretraining, long-context memory, and autoregressive video generation, all of which imply significant compute and data-infrastructure spend. | 中 | SI002 |
| CI013 | Rhoda’s research note says some tasks can be learned with roughly 10 hours of robot data, which if reproducible could reduce teleoperation expense relative to teleop-heavy competitors. | 高 | SI002, SI004 |
| CI014 | Public commercialization evidence is still framed as deployments and customer pilots rather than broad production fleets. | 中 | SI004, SI009, SI014 |
| CI015 | Business Wire, Yahoo Finance, Wilson Sonsini, and Rhoda’s own site corroborate the $450M March 2026 financing amount. | 高 | SI004, SI007, SI008, SI010 |
| CI016 | Secondary coverage consistently places the round at about a $1.7B valuation. | 中 | SI008, SI012, SI013, SI023, SI024 |
| CI017 | Rhoda’s own public materials do not identify a single lead investor even though several secondary outlets do. | 中 | SI004, SI007, SI009 |
| CI018 | Some secondary sources describe Premji Invest as the lead investor in the round. | 低 | SI014, SI024 |
| CI019 | The Wilson Sonsini transaction note instead describes the round as led by a multi-name syndicate. | 中 | SI010 |
| CI020 | Several third-party trackers or niche outlets introduce stage noise by classifying the 2026 financing inconsistently. | 低 | SI016, SI023 |
| CI021 | A SEC company search for “Rhoda AI” returned no matching companies. | 中 | SI017 |
| CI022 | A second SEC company search for “Rhoda Ai Corporation” also returned no matching companies. | 中 | SI018 |
| CI023 | California registry data shows Rhoda AI Corporation as a Delaware corporation incorporated on 2024-08-01 and active in California. | 中 | SI019 |
| CI024 | No public debt facilities, project-finance arrangements, or leverage disclosures appear in the retrieved source set. | 低 | SI004, SI017, SI018, SI019 |
| CI025 | Cash on hand and monthly burn are not publicly disclosed, so the public record cannot support a defensible runway calculation. | 中 | SI004, SI007, SI021 |
| CI026 | The 33-role hiring plan, especially across research, software, hardware, and operations, implies a materially expanding payroll base before public revenue proof. | 中 | SI020, SI025, SI026, SI027 |
| CI027 | Because Rhoda markets custom actuators, payload specs, and safety-rated vision, its cost structure likely includes hardware engineering and systems-integration expense on top of model training. | 中 | SI001, SI025, SI026 |
| CI028 | Because DVA relies on web-scale video pretraining and long-context video models, Rhoda also likely carries substantial compute and data-platform expense unlike a light software integrator. | 中 | SI002, SI027, SI015 |
| CI029 | The public record supports commercialization interest, but not revenue quality, because the strongest proof point is still a company-stated manufacturing benchmark plus pilot language. | 中 | SI004, SI011, SI021 |
| CI030 | futureTEKnow explicitly says Rhoda remains early in commercial rollout and still talks about industrial deployments and customer pilots rather than broad production fleets. | 中 | SI014 |
| CI031 | robotics.press argues Rhoda has zero independently validated deployments, zero named customers, and zero disclosed unit economics despite the $1.7B valuation. | 低 | SI021 |
| CI032 | The same robotics.press analysis says the only concrete operating KPI in public circulation comes from Rhoda’s own communications. | 低 | SI021 |
| CI033 | AgentMarketCap frames physical AI as a segment with expensive data collection, safety constraints, and deployment-specific integration, reinforcing Rhoda’s likely capital intensity. | 低 | SI015 |
| CI034 | TechStackIPO marks Rhoda as verification pending and includes an incorrect stage classification, which makes tracker-style financial metadata unsuitable for primary underwriting. | 低 | SI016 |
| CI035 | The Ashby base page shows every visible opening in Palo Alto, suggesting Rhoda’s current build-out is centered there rather than around a distributed field org. | 中 | SI020 |
| CI036 | The live job mix implies immediate spend on inference infrastructure, hardware leadership, supply chain, and operations rather than only research scientists. | 中 | SI025, SI026, SI027 |
| CI037 | Public materials do not disclose customer concentration, renewal rates, or contract duration, so revenue durability cannot be judged from outside the company. | 中 | SI004, SI007, SI021 |
| CI038 | Rhoda’s financing amount is unusually large for a first disclosed round, which reduces near-term fundraising pressure relative to most robotics startups at a similar public stage. | 中 | SI007, SI010, SI015 |
| CI039 | Even with $450M raised, the absence of disclosed burn means the next-round trigger cannot be underwritten from public data alone. | 中 | SI005, SI015, SI021 |
| CI040 | The public diligence verdict is therefore asymmetrical: strong capital base and credible technical ambition, but no public evidence yet for price realization, revenue quality, or margin path. | 中 | SI006, SI021, SI015 |
| CE001 | FutureVision is Rhoda’s intelligence layer and is intended to power Rhoda systems before expanding to partner hardware and software platforms. | 高 | SE001, SE003, SE010 |
| CE002 | Rhoda defines Direct Video Action as a policy where a causal video model predicts future frames and a separate inverse-dynamics model translates those predictions into robot actions in streaming closed loop. | 高 | SE002, SE003 |
| CE003 | Rhoda pre-trains its video model from scratch on general web videos rather than distilling from a pre-trained bidirectional model. | 中 | SE002 |
| CE004 | Rhoda says video-scale pretraining gives the model priors on 3D structure, physics, behavior and conventions. | 中 | SE002, SE005 |
| CE005 | Context Amortization predicts future video at every position in a long history so Rhoda can train causal video generation efficiently with hundreds of frames of context. | 中 | SE002 |
| CE006 | Rhoda’s Leapfrog Inference overlaps inference with action execution and conditions each new prediction on the action currently being executed to smooth trajectories. | 中 | SE002 |
| CE007 | Rhoda uses KV-caching at inference so encoded context can be reused across steps instead of recomputed from scratch. | 中 | SE002 |
| CE008 | The inverse-dynamics model performs video-to-action translation and Rhoda says it can be trained with as little as about 10 hours of embodiment data. | 高 | SE002, SE012 |
| CE009 | Rhoda says inverse-dynamics training can use random motions rather than only high-quality task demonstrations. | 中 | SE002 |
| CE010 | Rhoda reports that complex long-horizon tasks can be learned with 10–20 hours of robot data collected within a few days. | 高 | SE002, SE011, SE012 |
| CE011 | The bearing-decanting task used 11 hours of robot data and Rhoda says the system operated autonomously for 1.5 hours. | 中 | SE002 |
| CE012 | The Contico container-breakdown task used 17 hours of robot data and Rhoda says the system ran for 160 minutes continuously. | 中 | SE002, SE017 |
| CE013 | Rhoda says its models have hundreds of frames of visual context while many VLA systems operate with only a few frames. | 中 | SE002 |
| CE014 | The shell-game demo is meant to show persistent visual memory through multiple swaps of hidden objects. | 中 | SE002 |
| CE015 | Rhoda frames returns processing as an end-to-end workflow solved with long-context memory instead of hand-engineered progress indicators or multi-stage scaffolding. | 中 | SE001, SE002, SE016 |
| CE016 | Rhoda says one-shot pick-and-place and drawing demos use in-context learning from a single human demonstration without updating model weights. | 中 | SE001, SE002 |
| CE017 | Rhoda argues DVA improves interpretability because autoregressive video rollouts let engineers inspect model decisions, compare model variants and verify safe behavior. | 中 | SE002 |
| CE018 | Rhoda says a high-volume manufacturing workflow completed in under two minutes per cycle without human intervention during a customer evaluation. | 高 | SE003, SE010, SE012 |
| CE019 | Rhoda publicly names automotive, manufacturing, logistics and ecommerce as current commercial verticals. | 高 | SE001, SE003 |
| CE020 | Rhoda’s homepage advertises custom actuators with a 25 kg rated payload and 40 kg peak payload. | 中 | SE001 |
| CE021 | Rhoda’s homepage says the robot platform includes brakes in every actuator and safety-rated vision. | 中 | SE001 |
| CE022 | Rhoda’s homepage claims three years of continuous operation at rated payload. | 中 | SE001 |
| CE023 | Rhoda’s team and hiring materials describe the company as a full-stack effort spanning hardware, world models, cloud infrastructure, robot field operations and model training. | 中 | SE004, SE014, SE015 |
| CE024 | A Mayfield-hosted job posting says Rhoda’s cloud infrastructure supports data collection pipelines, robot operations and model training/evaluation workflows. | 中 | SE014 |
| CE025 | Careers-oriented sources place Rhoda in Palo Alto and show active hiring for infrastructure and robotics roles in 2026. | 中 | SE013, SE026 |
| CE026 | Independent commentary describes Rhoda as video-first and explicitly different from VLA systems that treat language as the primary control surface. | 中 | SE010, SE011 |
| CE027 | Mimic Robotics argues VLA backbones inherit semantics but not physical dynamics, which can make them less sample-efficient than video-model backbones. | 中 | SE023 |
| CE028 | The Kempner Institute argues web-scale video offers richer physical dynamics than static image-text pretraining for general-purpose robot planners. | 中 | SE022 |
| CE029 | GR00T N1 is an open VLA model rather than a causal-video policy and is built around multimodal inputs plus an action head. | 中 | SE018 |
| CE030 | GR00T N1.7 is commercially licensable under Apache 2.0 and pretrains on 20,000 hours of EgoScale human video, showing a more open benchmark path than Rhoda’s proprietary stack. | 中 | SE018 |
| CE031 | GR-2 illustrates that major rivals still keep language as a first-class control interface even when they add video generation and web-scale knowledge. | 中 | SE019 |
| CE032 | DreamGen shows that world-model approaches are converging on richer video pretraining to improve robot generalization outside narrow robot-demonstration corpora. | 中 | SE020 |
| CE033 | The 2025 robotics foundation-model review says safety, data diversity, embodiment and compute remain unresolved bottlenecks across the category. | 中 | SE021 |
| CE034 | Coey argues Rhoda’s current public evidence is still demo-led and not standardized by third-party benchmarking. | 低 | SE011 |
| CE035 | No public third-party safety certification, formal audit or standardized benchmark for Rhoda’s DVA stack was found in the reviewed sources. | 中 | SE001, SE002, SE003, SE011 |
| CE036 | No public SDK, API documentation or developer repository was found in the reviewed official materials, so Rhoda’s current developer signal is hiring-oriented rather than ecosystem-oriented. | 低 | SE001, SE004, SE013, SE014 |
| CE037 | Google DeepMind’s Gemini Robotics claims a different VLA path to robustness, with benchmark and embodiment claims that increase competitive pressure on Rhoda’s narrative. | 中 | SE025 |
| CE038 | Rhoda’s official videos and blog show production-style demos, but public sources still lack standardized customer-by-customer pass/fail distributions or uptime cohorts. | 中 | SE002, SE016, SE017, SE011 |
| CU001 | Rhoda’s homepage says the company works with customers across automotive, manufacturing, logistics, and ecommerce. | 中 | SU001 |
| CU002 | Rhoda’s public customer surface is workflow-based rather than account-based, centering on returns processing, bearing decanting, and Contico breakdown. | 中 | SU001 |
| CU003 | Rhoda describes returns processing as an end-to-end task for a customer in the logistics industry. | 中 | SU001 |
| CU004 | Rhoda describes bearing decanting as a task from an automotive assembly line. | 中 | SU001 |
| CU005 | Rhoda describes Contico breakdown as a manufacturing workflow involving 50-pound heavy-duty boxes used to move materials between facilities. | 中 | SU001 |
| CU006 | None of Rhoda’s reviewed official customer materials publicly names the logistics customer shown in returns processing. | 高 | SU001, SU002, SU010 |
| CU007 | None of Rhoda’s reviewed official customer materials publicly names the automotive assembly customer or the high-volume manufacturing evaluation customer. | 高 | SU001, SU002, SU010 |
| CU008 | Rhoda’s press release says the company works with leading industrial partners across manufacturing and logistics. | 高 | SU002, SU003, SU006 |
| CU009 | Rhoda says its technology has already demonstrated autonomous operation in production environments. | 高 | SU002, SU003, SU006 |
| CU010 | Rhoda says a recent high-volume manufacturing evaluation completed a component-processing workflow in under two minutes per cycle without human intervention. | 高 | SU002, SU003, SU006, SU014 |
| CU011 | Rhoda says FutureVision is expected over time to be licensed to partners across different robotic hardware and software platforms. | 高 | SU002, SU003, SU005, SU013 |
| CU012 | Reuters reported that Rhoda’s platform is designed to integrate with a wide range of robotic hardware so manufacturers and logistics operators can deploy intelligent robots without rebuilding existing systems. | 中 | SU013 |
| CU013 | RoboHorizon framed Rhoda as a hardware-agnostic intelligence layer that could upgrade existing fleets of robots. | 中 | SU008, SU025 |
| CU014 | Rhoda’s press release says the March 2026 financing will support expansion of industrial deployments and customer pilots. | 高 | SU002, SU003, SU004, SU015 |
| CU015 | Wilson Sonsini separately described the funding use as expansion of industrial deployments and customer pilots. | 中 | SU015 |
| CU016 | Rhoda’s Direct Video-Action research blog says its model can robustly learn real-world long-horizon tasks with roughly 10–20 hours of robot data. | 中 | SU010 |
| CU017 | Rhoda’s Direct Video-Action research blog says two example customer tasks were deployed as real customer proof of concepts and operated successfully for multiple hours without human intervention. | 中 | SU010 |
| CU018 | Rhoda says the decanting task reached autonomous operation after 11 hours of task data and ran for 1.5 hours in one uncut demonstration. | 中 | SU010 |
| CU019 | Rhoda says the container breakdown task reached a high degree of robustness after 17 hours of robot data and ran for 160 minutes in one continuous demonstration. | 中 | SU010 |
| CU020 | Rhoda says long-context memory lets its returns-processing workflow run end to end without hand-engineered task scaffolding. | 中 | SU010 |
| CU021 | Rhoda’s homepage says long-context memory enables single-shot learning from human demonstrations. | 高 | SU001, SU010 |
| CU022 | Humanoids Daily reported that Rhoda demonstrated hardware inside one of the world’s largest automotive factories. | 中 | SU007 |
| CU023 | Humanoids Daily reported that Rhoda’s decanting workflow handled 10 kg boxes, straps, tabs, and deformable bags in a live automotive setting. | 中 | SU007 |
| CU024 | Rhoda’s public materials do not disclose customer count, pipeline size, NRR, GRR, churn, contract length, or renewal rates. | 高 | SU001, SU002, SU010, SU011, SU012 |
| CU025 | Rhoda’s public materials do not disclose customer ROI, labor-savings, or payback metrics. | 高 | SU001, SU002, SU010 |
| CU026 | NIST says only 10% of potential manufacturing users have adopted robotic systems because buyers still lack assurance that systems can be readily integrated and will perform under dynamic shop-floor conditions. | 中 | SU016 |
| CU027 | NIST says lengthy and expensive installation plus missing metrics, benchmarks, and interoperability infrastructure remain barriers to broader manufacturing robot adoption. | 中 | SU016 |
| CU028 | Automate reported that scaling custom-coded robot solutions across facilities becomes prohibitively expensive for enterprise customers. | 中 | SU017 |
| CU029 | Automate reported that exceptions and poor machine-to-machine communication can halt production and turn facility automation into a system that works until it does not. | 中 | SU017 |
| CU030 | MDPI’s 2025 review said high implementation costs and legacy-system incompatibilities still hinder industrial robot adoption, especially for SMEs. | 中 | SU018 |
| CU031 | MDPI’s 2025 review said interoperability gaps, workforce displacement concerns, and cybersecurity risks remain unresolved in industrial robotics. | 中 | SU018 |
| CU032 | PwC says supply chains now face severe material, energy, and talent shortages, which is pushing operators toward more cognitive and adaptable systems. | 中 | SU019 |
| CU033 | Mordor Intelligence says the warehouse automation market is expected to grow from USD 29.98 billion in 2025 to USD 34.17 billion in 2026, supported by labor shortages, wage inflation, rapid ROI from plug-and-play robotics, and Robotics-as-a-Service models. | 中 | SU020 |
| CU034 | Mordor Intelligence says returns processing is among the faster-growing warehouse automation functions through 2031. | 中 | SU020 |
| CU035 | MarketsandMarkets says long commercialization timelines and high maintenance costs are explicit challenges in the AI robots market. | 中 | SU021 |
| CU036 | MarketsandMarkets says reluctance to adopt new technologies and the absence of standardized regulations remain important adoption restraints for AI robots. | 中 | SU021 |
| CU037 | Monocle argued that only about 10% of companies sustain large-scale warehouse automation success beyond pilots because ROI models often understate integration, downtime, and change-management costs. | 低 | SU023 |
| CU038 | Because Rhoda’s public proof is workflow-level and unnamed, concentration risk and referenceability risk are higher than the technical demos alone suggest. | 中 | SU001, SU002, SU010, SU023 |
| CU039 | Amazon said it licensed Covariant’s robotic foundation models and hired Covariant’s founders to accelerate intelligent and safe warehouse robotics at scale. | 中 | SU026 |
| CU040 | KNAPP said it extended its success story with Covariant, showing that warehouse-automation partners in this category sometimes publicly disclose ongoing robotics-AI relationships. | 中 | SU027 |
| CU041 | Modern Materials Handling reported that GXO piloted Dexterity’s AI-enhanced robotics in warehouse operations, providing a named-pilot reference point that buyers and investors can scrutinize. | 中 | SU028 |
| CU042 | Relative to named reference points disclosed by Amazon-Covariant, KNAPP-Covariant, and GXO-Dexterity, Rhoda’s still-unnamed workflow evidence leaves a larger customer-validation gap in the public record. | 中 | SU001, SU002, SU026, SU027, SU028 |
| CR001 | Rhoda’s public launch materials frame the company around bringing robots from controlled lab demos into real-world industrial environments. | 高 | SR017, SR024, SR021 |
| CR002 | Rhoda’s reviewed official pages do not publicly name customers, customer count, or deployment count. | 高 | SR016, SR017, SR018, SR019, SR020 |
| CR003 | Rhoda’s reviewed official pages do not publish safety certifications, compliance pages, incident metrics, or recall disclosures. | 高 | SR016, SR017, SR018, SR019, SR020 |
| CR004 | Rhoda says Direct Video Action models use internet-scale video pretraining and then smaller amounts of robot data to learn embodiment-specific behaviors. | 高 | SR017, SR018, SR024 |
| CR005 | Rhoda says its model often requires as little as ten hours of teleoperation data to learn new tasks efficiently. | 高 | SR017, SR021 |
| CR006 | Rhoda’s research blog says its model can learn real-world long-horizon tasks with roughly 10–20 hours of robot data. | 中 | SR018 |
| CR007 | Rhoda’s research blog says two example customer tasks were real customer proof-of-concepts that operated for multiple hours without human intervention. | 中 | SR018 |
| CR008 | Rhoda’s research blog presents interpretability through video generation as a way to inspect model behavior and compare configurations. | 中 | SR018 |
| CR009 | Interpretability through generated video is helpful for debugging, but it is not a substitute for public safety certification, incident reporting, or deployment-grade reliability metrics. | 中 | SR018, SR025 |
| CR010 | Reuters reported that reliability, safety certification, and cost remain key hurdles for large-scale commercial deployment of general-purpose robots. | 中 | SR025 |
| CR011 | OSHA says many robot accidents occur during non-routine conditions such as programming, maintenance, testing, setup, or adjustment. | 中 | SR001 |
| CR012 | OSHA says there are currently no specific OSHA standards for the robotics industry. | 中 | SR001 |
| CR013 | NIOSH says its Center for Occupational Robotics Research monitors injury trends, evaluates robotics technologies, and supports the development of consensus safety standards. | 中 | SR002 |
| CR014 | The EU AI Act says AI may generate physical, psychological, societal, or economic harm and sets uniform obligations to protect health, safety, and fundamental rights. | 中 | SR003 |
| CR015 | The European Commission says the new Machinery Regulation integrates provisions for AI-powered safety functions and cyber-safety and applies on a mandatory basis from 20 January 2027. | 中 | SR004 |
| CR016 | ISO 10218-1 specifies safe-design requirements and protective measures for industrial robots. | 中 | SR005 |
| CR017 | ISO 10218-2 covers robot systems and integration, and ISO/TS 15066 covers collaborative-robot operation. | 高 | SR006, SR007 |
| CR018 | Harvard JOLT argues that mere human presence is insufficient oversight for high-risk AI systems and that deployers and developers need real collaboration frameworks, technical robustness, and post-market monitoring. | 中 | SR008 |
| CR019 | Brookings says AI harms can trigger negligence, design-defect, failure-to-warn, and other products-liability theories, and companies cannot legitimately blame the AI itself when foreseeable use causes harm. | 中 | SR009 |
| CR020 | NIST says only about 10% of potential manufacturing users have adopted robotic systems because they still lack assurance around integration and performance under dynamic shop-floor conditions. | 中 | SR010 |
| CR021 | NIST says gaps in metrics, benchmarks, and standards hinder the transition of research breakthroughs into commercially available industrial robots. | 中 | SR010 |
| CR022 | Automate reported that interoperability problems, exception handling, and poor machine-to-machine communication can halt factories and make scaling custom-coded robot solutions prohibitively expensive. | 中 | SR011 |
| CR023 | IndustrialEngineer.ai argued that warehouse robotics frequently miss ROI targets when they are bolted onto broken WMS and labor processes. | 低 | SR012 |
| CR024 | Cleverence’s 3PL case study says robotics projects rise or fall on change management, integration quality, reporting, and tariff discipline rather than hardware alone. | 低 | SR013 |
| CR025 | Rhoda investor messaging says the first company to deploy intelligent manipulation-capable robots at scale can build a compounding data flywheel from real-world edge cases. | 高 | SR017, SR024 |
| CR026 | Because Rhoda does not publicly disclose named customers or deployment counts, the existence of a scaled data-flywheel moat is not externally verifiable today. | 中 | SR016, SR017, SR018, SR025 |
| CR027 | Reuters says Rhoda’s platform is designed to integrate with a wide range of robotic hardware so manufacturers and logistics operators can deploy intelligent robots without rebuilding existing systems. | 中 | SR025 |
| CR028 | A hardware-agnostic commercialization model shifts execution risk toward third-party robot hardware, software stacks, and on-site integrators that Rhoda does not publicly enumerate. | 中 | SR017, SR023, SR025, SR030 |
| CR029 | Rhoda’s public leadership page names research and product leaders, but the public site does not disclose dedicated safety, compliance, or field-operations functions. | 中 | SR019, SR020 |
| CR030 | Humanoids Daily reported that Rhoda plans not only to license software but also to develop its own hardware as a data-collection engine. | 中 | SR022 |
| CR031 | Humanoids Daily described Rhoda’s March 2026 funding as a Series B round. | 中 | SR022 |
| CR032 | Rhoda’s official materials, Reuters, Tech Funding News, and Robotics & Automation News describe the March 2026 financing as a Series A round. | 高 | SR017, SR021, SR025, SR026, SR027 |
| CR033 | Humanoids Daily reported that Rhoda demonstrated operation in one of the world’s largest automotive factories. | 中 | SR022 |
| CR034 | CNBC, citing Barclays, said the humanoid market is only about $2 billion to $3 billion today but could reach $200 billion by 2035, with China already dominating installations and production cost. | 中 | SR014 |
| CR035 | CNBC also noted that risks around robots will need to be carefully balanced by industry and governments even as productivity expectations rise. | 中 | SR014 |
| CR036 | Humanoids Daily’s competition article said Rhoda and Genesis entered a crowded field alongside Figure, Tesla, and 1X, and quoted investor caution that commercialization remains long and fraught with engineering challenges. | 中 | SR015 |
| CR037 | RoboHorizon said Rhoda is positioning itself as a brains provider for the broader industrial market, which increases upside but also dependence on other vendors’ physical platforms. | 中 | SR023, SR030 |
| CR038 | Rhoda’s official pages do not publish a post-market monitoring, failure-reporting, or incident-response framework for deployed systems. | 高 | SR016, SR017, SR018 |
| CR039 | The public materials say Rhoda has industrial deployments and customer pilots, but they do not disclose the denominator needed to judge conversion, concentration, or site-level durability. | 高 | SR017, SR021, SR026, SR027 |
| CR040 | European commercialization would likely require Rhoda and its partners to navigate both AI-system obligations and machinery conformity processes before broad deployment. | 高 | SR003, SR004, SR005, SR006, SR007 |
| CR041 | Rhoda’s strongest public commercial evidence is still a combination of unnamed industrial workflows, customer proof-of-concepts, and one quantified manufacturing evaluation rather than a named installed base. | 高 | SR016, SR017, SR018, SR021, SR025 |
| CR042 | A material de-rating trigger would be any evidence that Rhoda’s unnamed evaluations fail to convert into referenceable deployments while the broader physical-AI market stays crowded and hype-heavy. | 中 | SR014, SR015, SR017, SR021 |
| CR043 | No reviewed public source disclosed customer incidents, recalls, insurance details, or liability coverage specific to Rhoda’s deployments. | 高 | SR016, SR017, SR018, SR025 |
| CV001 | Rhoda publicly announced a $450 million Series A on 2026-03-10 after 18 months in stealth. | 高 | SV001, SV002, SV003, SV004 |
| CV002 | Multiple secondary outlets cited a Rhoda valuation of about $1.7 billion for the March 2026 round. | 中 | SV005, SV006, SV007 |
| CV003 | Rhoda presents FutureVision as a licensable intelligence layer intended to work across robotic hardware and software platforms. | 高 | SV001, SV002 |
| CV004 | Rhoda says it has already demonstrated autonomous operation in production environments and exceeded customer KPIs in a manufacturing evaluation. | 中 | SV002, SV003 |
| CV005 | Rhoda’s public materials do not disclose revenue, gross margin, pricing, or named paying customers. | 中 | SV001, SV002, SV003, SV004 |
| CV006 | Humanoids Daily described Rhoda’s March 2026 round as a Series B, while Rhoda and Wilson Sonsini describe it as a Series A. | 中 | SV001, SV004, SV006 |
| CV007 | AgentMarketCap says 27 physical-AI startups raised more than $6 billion in Q1 2026, including roughly $4 billion for robotics companies. | 中 | SV005 |
| CV008 | Skild AI raised a $1.4 billion Series C in January 2026 at a valuation above $14 billion. | 高 | SV008, SV009 |
| CV009 | Skild AI said it grew from zero to about $30 million of revenue in just a few months in 2025. | 中 | SV009 |
| CV010 | Skild’s disclosed $14 billion valuation and about $30 million revenue imply a trailing revenue multiple of roughly 467x. | 中 | SV008, SV009 |
| CV011 | Physical Intelligence was reportedly in talks in March 2026 to raise about $1 billion at a valuation exceeding $11 billion. | 中 | SV010 |
| CV012 | TechCrunch reported that Physical Intelligence had no timeline for commercialization despite investor appetite for a larger round. | 中 | SV010 |
| CV013 | The Robot Report said Physical Intelligence raised $600 million in late 2025 and was valued at about $5.6 billion according to Bloomberg. | 中 | SV016 |
| CV014 | Figure raised a Series C round in September 2025 that valued it at $39 billion. | 中 | SV011, SV012 |
| CV015 | Sacra says Figure’s September 2025 valuation represented about a 15x increase from its $2.6 billion Series B valuation in February 2024. | 中 | SV012 |
| CV016 | TechCrunch reported Dexterity raised $95 million at a $1.65 billion post-money valuation in March 2025. | 中 | SV013 |
| CV017 | FieldAI disclosed $405 million of total funding in August 2025 without publicly disclosing a valuation. | 中 | SV014 |
| CV018 | CNBC reported Apptronik raised $520 million at a $5 billion valuation in February 2026. | 中 | SV015 |
| CV019 | The private physical-AI comparable ladder currently runs from about $1.65 billion for Dexterity to about $39 billion for Figure. | 中 | SV011, SV013, SV015, SV016 |
| CV020 | Rhoda’s roughly $1.7 billion mark sits near Dexterity’s level and well below Apptronik, Physical Intelligence, Skild, and Figure. | 中 | SV005, SV007, SV008, SV010, SV011, SV013, SV015 |
| CV021 | Symbotic’s 2025 10-K says the company had about $22.5 billion of backlog as of September 27, 2025. | 中 | SV018 |
| CV022 | Yahoo Finance listed Symbotic at roughly $6.03 billion market cap, 2.27x price/sales, and $2.52 billion trailing revenue as of 2026-06-04. | 高 | SV018, SV019 |
| CV023 | Yahoo Finance listed Zebra at roughly $11.06 billion market cap, 2.10x price/sales, and $5.58 billion trailing revenue as of 2026-06-05. | 中 | SV020, SV023 |
| CV024 | Yahoo Finance listed Rockwell at roughly $49.71 billion market cap, 5.72x price/sales, and $8.8 billion trailing revenue as of 2026-06-05. | 中 | SV021, SV022 |
| CV025 | Eilla’s robotics valuation playbook says warehouse and intralogistics robotics systems or RaaS businesses often trade around 2.2x-5.0x EV/revenue in precedent analysis. | 中 | SV024 |
| CV026 | Eilla says services-heavy industrial automation integrators often screen lower, around 0.9x-2.1x EV/revenue. | 中 | SV024 |
| CV027 | Rhoda cannot be translated into a defensible observed revenue multiple because public evidence provides no revenue denominator. | 中 | SV001, SV002, SV003, SV004, SV005, SV006, SV007 |
| CV028 | Rhoda’s $1.7 billion price is therefore underwritten mainly on option value, team quality, and category narrative rather than on disclosed fundamentals. | 中 | SV002, SV004, SV005, SV024, SV027, SV028 |
| CV029 | McKinsey says logistics and fulfillment players expect automation to represent 30% or more of capital spending over the next five years. | 中 | SV029 |
| CV030 | McKinsey says 71% of industrial respondents cite capital cost and 61% cite lack of automation experience as adoption barriers. | 中 | SV029 |
| CV031 | The U.S. Chamber says durable-goods manufacturing still had about 313,000 open jobs in April 2025. | 中 | SV030 |
| CV032 | BLS projects about 1.008 million annual openings for hand laborers and material movers, preserving a large automation opportunity pool. | 中 | SV031 |
| CV033 | NVIDIA says 90% of surveyed retail and CPG respondents plan to increase AI budgets in 2026. | 中 | SV027 |
| CV034 | UPS says logistics technology investment in 2026 is shifting toward resilience, AI, robotics, software-defined warehouses, and RaaS. | 中 | SV028 |
| CV035 | A bear-case valuation for Rhoda of about $0.9 billion to $1.3 billion is plausible if investors anchor more tightly to public automation multiples and commercialization remains opaque. | 中 | SV019, SV020, SV021, SV024, SV025 |
| CV036 | A base-case valuation band of about $1.5 billion to $2.2 billion fits the current mark only if pilots convert into clearer recurring software revenue and named customer proof emerges. | 中 | SV002, SV005, SV024, SV028, SV029 |
| CV037 | A bull-case valuation band of about $3 billion to $5 billion would require Rhoda to prove platform licensing, durable customer expansion, and materially better evidence than is public today. | 中 | SV008, SV010, SV011, SV012, SV015 |
| CV038 | At public-style revenue multiples, Rhoda would need roughly $340 million of revenue to justify $1.7 billion at 5x sales and roughly $170 million at 10x sales. | 中 | SV020, SV021, SV024 |
| CV039 | Rhoda’s current price looks less extreme than Skild, Figure, or Physical Intelligence on an absolute basis, but still aggressive for a company with undisclosed revenue and customer economics. | 中 | SV005, SV008, SV010, SV011, SV024 |
| CV040 | The recommendation implied by current public evidence is TRACK or research-more rather than aggressive buy at a higher step-up from today’s mark. | 中 | SV005, SV024, SV028, SV029 |
| CV041 | A thesis-break trigger is failure to disclose named paid deployments, pricing, or repeatable revenue conversion by the next financing cycle. | 中 | SV002, SV005, SV028 |
| CV042 | A second thesis-break trigger is a sector reset in which better-proven peers raise flat or down rounds, making Rhoda’s narrative premium harder to defend. | 中 | SV010, SV011, SV013, SV015, SV024 |
| CV043 | FieldAI’s undisclosed valuation and Rhoda’s undisclosed revenue both illustrate how much of the current physical-AI market still relies on opaque private marks rather than auditable denominators. | 中 | SV005, SV014 |
| CV044 | Figure’s and Skild’s far larger valuations reflect stronger public fundraising scale and, in Figure’s case, explicit large-customer deployment reporting that Rhoda has not yet matched publicly. | 中 | SV008, SV009, SV011, SV012 |
| CV045 | Rhoda’s team pedigree is a positive signal, but public evidence still leaves monetization mechanics, gross margin, and customer concentration unresolved. | 中 | SV002, SV004, SV026 |
| 编号 | 出版方 | 标题 | 引文 |
|---|---|---|---|
| SO001 | Rhoda AI | Rhoda AI | FutureVision brings the capability to handle real world industrial tasks autonomously. |
| SO002 | Rhoda AI | Causal Video Models Are Data-Efficient Robot Policy Learners | Rhoda AI | Our models perform complex, long-horizon tasks reliably with as little as ~10 hours of total robot data. |
| SO003 | Rhoda AI | News | Rhoda AI | |
| SO004 | Rhoda AI | Press Release | Rhoda AI | Rhoda AI today announced its public launch after 18 months in stealth. |
| SO005 | Rhoda AI | Team | Rhoda AI | |
| SO006 | Rhoda AI | Careers | Rhoda AI | |
| SO007 | Rhoda AI | Contact | Rhoda AI | |
| SO008 | Ashby | Rhoda AI Jobs | Open Positions (33) |
| SO009 | Business Wire | Rhoda AI Exits Stealth with $450 Million Series A to Bring Robots Out of the Lab and Into the Real World | |
| SO010 | Yahoo Finance | Rhoda AI raises $450 million to accelerate industrial deployment | |
| SO011 | TechNode Global | Temasek-backed Rhoda AI raises $450M Series A funding to accelerate robotics development | |
| SO012 | Wilson Sonsini | Wilson Sonsini Advises Rhoda AI on $450 Million Funding as Company Emerges from Stealth | |
| SO013 | RoboticsTomorrow | Rhoda AI Exits Stealth with $450 Million Series A to Bring Robots Out of the Lab and Into the Real World | RoboticsTomorrow | |
| SO014 | Robotics & Automation News | Rhoda AI raises $450 million to develop real-world robotic intelligence | |
| SO015 | Tech Funding News | Khosla-backed Rhoda raises $450M at $1.7B valuation for video-trained AI | |
| SO016 | Humanoids Daily | Rhoda AI Hits $1.7B Valuation, Unveils "Direct Video-Action" Model to Bridge the Real-World Gap | |
| SO017 | RoboHorizon | Rhoda AI Unveils Video-Trained Robots, Nabs $450M at $1.7B Valuation | |
| SO018 | intelligence360 | Rhoda AI Exits Stealth with $450 Million Series A to Bring Robots Out of the Lab and Into the Real World | |
| SO019 | US Finance Insider | AI Robotics Startup Rhoda Hits US$1.7 Billion Valuation after Successful Funding Round | |
| SO020 | Stanford University | Gordon Wetzstein - Stanford University | since 10/24 Co-founder, Rhoda AI |
| SO021 | California Companies Directory | Rhoda AI Corporation | California Companies Direcotry | Rhoda Ai Corporation was incorporated as Stock Corporation on 1 August 2024. |
| SO022 | Tracxn | Rhoda - 2026 Company Profile & Team - Tracxn | Rhoda has 60 employees as of Mar 26. |
| SO023 | Prelude Ventures | Rhoda AI | |
| SO024 | Creati.ai | Rhoda AI Raises $450 Million at $1.7 Billion Valuation to Train Robots Using Internet Videos | |
| SO025 | robotics.press | Rhoda AI | robotics.press | |
| SO026 | TechStackIPO | Rhoda AI — Funding, Valuation & IPO Status | |
| SM001 | Rhoda AI | News | Rhoda AI | Rhoda AI today announced its public launch after 18 months in stealth, unveiling FutureVision, a new approach to robotic intelligence based on video-predictive control. |
| SM002 | Rhoda AI | Team | Rhoda AI | Our Leadership Team ... Jagdeep Singh ... Eric Chan ... Gordon Wetzstein ... and a team drawn from leading generative AI, computer vision, and robotics organizations. |
| SM003 | Business Wire | Rhoda AI Exits Stealth with $450 Million Series A to Bring Robots Out of the Lab and Into the Real World | FutureVision serves as Rhoda’s intelligence layer — a foundation model that powers Rhoda systems today and is expected over time to be licensed to partners across different robotic hardware and software platforms. |
| SM004 | TNGlobal | Temasek-backed Rhoda AI raises $450M Series A funding to accelerate robotics development | The Series A will support its continued research and engineering investment, expansion of industrial deployments and customer pilots. |
| SM005 | Robotics & Automation News | Rhoda AI launches with $450 million Series A to bring robots out of the lab and into the real world | Rhoda’s technology has already demonstrated autonomous operation in production environments, where robots must handle continuously changing materials, layouts, and workflows. |
| SM006 | Wilson Sonsini | Wilson Sonsini Advises Rhoda AI on $450 Million Funding as Company Emerges from Stealth | Rhoda AI has developed new technology for efficiently training robots to handle real world industrial tasks autonomously. |
| SM007 | MarketsandMarkets | Artificial Intelligence (AI) Robots Market Report 2025 - 2030 | The global artificial intelligence robots market is projected to grow from USD 6.11 billion in 2025 to USD 33.39 billion by 2030, at a CAGR of 40.4%. |
| SM008 | Mordor Intelligence | Warehouse Automation Market - Industry Size & Growth 2025 - 2031 | The Warehouse Automation Market worth USD 34.17 billion in 2026 is growing at a CAGR of 13.98% to reach USD 65.74 billion by 2031. |
| SM009 | International Federation of Robotics | World Robotics 2025 report – Industrial Robots – released by IFR | The total number of industrial robots in operational use worldwide was 4,664,000 units in 2024 – an increase of 9% compared to the previous year. |
| SM010 | International Federation of Robotics | Collaborative Robots - How Robots Work alongside Humans | Cobots accounted for 10.5% of the total 541,302 industrial robots installed in 2023. |
| SM011 | McKinsey & Company | Automation in logistics: Big opportunity, bigger uncertainty | With all this complexity comes a lot of uncertainty: Where should new fulfillment centers be built? ... How much and what kind of automation is needed? |
| SM012 | McKinsey & Company | Unlocking the industrial potential of robotics and automation | For logistics and fulfillment players, automation will represent 30 percent or more of their capital spending in the next five years. |
| SM013 | StartUs Insights | Third Party Logistics Report 2026 [Free PDF] | The global third-party logistics (3PL) market is projected to grow from USD 1.8 trillion in 2026 to USD 4.3 trillion by 2035 at a compound annual growth rate (CAGR) of 10.1%. |
| SM014 | U.S. Chamber of Commerce | Understanding America’s Labor Shortage: The Most Impacted Industries | As of April 2025, a gap persists, with 313,000 durable goods manufacturing job openings yet to be filled. |
| SM015 | Bureau of Labor Statistics | Hand Laborers and Material Movers | About 1,008,300 openings for hand laborers and material movers are projected each year, on average, over the decade. |
| SM016 | Hy-Tek Intralogistics | 2026 Warehouse Automation Trends: Where Software, AI, and Robotics Converge | What used to be a hardware-driven industry is now powered by software intelligence, artificial intelligence (AI), and robotics that work together to deliver unprecedented agility and throughput. |
| SM017 | Modern Materials Handling | 2026 Automation Study: Warehouse automation ticks upward | Global organizations invested about $21 billion in warehouse automation in 2023 ... By 2033, that number is expected to exceed $90 billion. |
| SM018 | Automated Warehouse / Interact Analysis | Warehouse automation starts 2025 strong, but faces uncertainty, says Interact Analysis | Warehouse automation forecasts have been revised down due to slow growth in the mobile robot segment. |
| SM019 | Automated Warehouse / Interact Analysis | Interact Analysis sees uncertainty for warehouse automation in 2026 | Warehouse automation revenue grew by 1%, compared with the -3% decline we had previously predicted. |
| SM020 | DHL | Logistics Industry Trends for 2026 | In 2026, AI will handle routine but essential tasks on its own ... and more SMEs will lean on smart systems that automatically move stock, vehicles, and people to where they’re needed most. |
| SM021 | UPS Supply Chain Solutions | 2026 Supply Chain Outlook | Logistics technology investment is accelerating ... Robotics and autonomous mobile robots improving warehouse productivity and accuracy ... Software-defined warehouses integrating enterprise systems, robotics and real-time data. |
| SM022 | NVIDIA | From Warehouse to Wallet: New State of AI in Retail and CPG Survey Uncovers How AI Is Rewiring Supply Chains and Customer Experiences | 90% said they’d build on the success of current projects by increasing their AI budgets in 2026. |
| SM023 | The Robot Report | Skild AI grabs $300M to build foundation model for robotics | With a horizontal market approach, you create a broadly intelligent system that is capable of learning any task and then make it capable of being deployed to control any mechanism. |
| SM024 | The Robot Report | Physical Intelligence raises $600M to advance robot foundation models | Other companies are also racing to get the data and build the models for next-generation robot AI. |
| SM025 | MarketsandMarkets | Physical AI Market Size, Share, Growth & Trends by Offering ... Global Forecast to 2032 | The global physical AI market Size is projected to grow from USD 1.50 billion in 2026 to USD 15.24 billion by 2032 at a CAGR of 47.2%. |
| SP001 | Rhoda AI | Press Release | Rhoda AI | FutureVision serves as Rhoda’s intelligence layer — a foundation model that powers Rhoda systems today and is expected over time to be licensed to partners across different robotic hardware and software platforms. |
| SP002 | Rhoda AI | Causal Video Models Are Data-Efficient Robot Policy Learners | At Rhoda AI, we are building towards generalist robotics. Our Direct Video-Action Model (DVA) reformulates robot policies as video generation. |
| SP003 | Skild AI | Skild AI | |
| SP004 | BusinessWire | Skild AI Raises $1.4B, Now Valued Over $14B | |
| SP005 | Automate | Skild.AI is Tackling the Physical AI Data Gap with $1.4B in New Funds | Learning a common model across different form factors is a necessity. |
| SP006 | Physical Intelligence | Our First Generalist Policy | Our first step is π0, a prototype model that combines large-scale multi-task and multi-robot data collection with a new network architecture. |
| SP007 | GitHub | GitHub - Physical-Intelligence/openpi | openpi holds open-source models and packages for robotics, published by the Physical Intelligence team. |
| SP008 | arXiv | $π_0$: A Vision-Language-Action Flow Model for General Robot Control | We propose a novel flow matching architecture built on top of a pre-trained vision-language model (VLM). |
| SP009 | Figure | Helix | Figure | Helix is designed to reason like a human. |
| SP010 | Humanoid Index | Figure AI: Funding, Valuation, Robot Specs & More | Figure 02 humanoid robot. BMW pilot deployment. $39B valuation — highest in humanoid robotics. |
| SP011 | TechMarketBriefs | Figure AI IPO 2026: $39B Valuation, Risks & Bull Case | The bear case is everything else: a valuation roughly equal to Goldman’s projected 2035 humanoid TAM. |
| SP012 | Dexterity | Dexterity - Physical AI | Our world model for Physical AI - trained with experience from over 100 million autonomous actions in production. |
| SP013 | Field AI | Redefining Industrial AI | Leading the frontier of Physical AI with deployments across three continents. |
| SP014 | Apptronik | Apptronik | |
| SP015 | Apptronik | Apollo | Apollo is the first commercial humanoid robot that was designed for friendly interaction, mass manufacturability, high payloads and safety |
| SP016 | GitHub | GitHub - NVIDIA/Isaac-GR00T | GR00T N1.7 is fully commercially licensable under Apache 2.0. |
| SP017 | NVIDIA | NVIDIA and Global Robotics Leaders Take Physical AI to the Real World | Leading developers such as FieldAI and Skild AI are building generalized robot brains using NVIDIA Cosmos world models and Isaac simulation frameworks. |
| SP018 | Google DeepMind | Introducing Gemini Robotics and Gemini Robotics-ER, AI models designed for robots to understand, act and react to the physical world. | Gemini Robotics is an advanced vision-language-action (VLA) model. |
| SP019 | Google DeepMind | Gemini 3.5 | |
| SP020 | Physical Intelligence | Physical Intelligence (π) | |
| SP021 | EVS | Top Robotics Foundation Model & Embodied AI Companies 2026 | |
| SP022 | Raise Summit | 20 Physical AI Companies to Watch in 2026 | |
| SP023 | Standard Bots | Top AI robotics companies to watch in 2026 (and what they’re actually building) | |
| SP024 | SiliconANGLE | Robot software startup Skild AI raises $1.4B round backed by Nvidia, Jeff Bezos | |
| SP025 | AI2Work | Skild AI’s $1.4B Raise: Why Robotics Foundation Models Are 2026’s Mega-Bet | |
| SP026 | Edge AI and Vision Alliance | NVIDIA and Global Robotics Leaders Take Physical AI to the Real World | |
| SP027 | Covariant | Covariant | |
| SI001 | Rhoda AI | Rhoda AI | |
| SI002 | Rhoda AI | Causal Video Models Are Data-Efficient Robot Policy Learners | Rhoda AI | |
| SI003 | Rhoda AI | News | Rhoda AI | |
| SI004 | Rhoda AI | Press Release | Rhoda AI | |
| SI005 | Rhoda AI | Team | Rhoda AI | |
| SI006 | Ashby | Rhoda AI embed script | |
| SI007 | Business Wire | Rhoda AI Exits Stealth with $450 Million Series A to Bring Robots Out of the Lab and Into the Real World | |
| SI008 | Yahoo Finance | Rhoda AI raises $450 million to accelerate industrial deployment | |
| SI009 | TechNode Global | Temasek-backed Rhoda AI raises $450M Series A funding to accelerate robotics development | |
| SI010 | Wilson Sonsini | Wilson Sonsini Advises Rhoda AI on $450 Million Funding as Company Emerges from Stealth | |
| SI011 | Robotics & Automation News | Rhoda AI raises $450 million to develop real-world robotic intelligence | |
| SI012 | Tech Funding News | Khosla-backed Rhoda raises $450M at $1.7B valuation for video-trained AI | |
| SI013 | RoboHorizon | Rhoda AI Unveils Video-Trained Robots, Nabs $450M at $1.7B Valuation | |
| SI014 | futureTEKnow | Rhoda AI robot intelligence hits $1.7B | Despite the strong capital backing, the company remains early in commercial rollout, with references to “industrial deployments and customer pilots” rather than broad production fleets. |
| SI015 | AgentMarketCap | Rhoda AI's $450M Series A Signals the Physical AI Agent Boom | |
| SI016 | TechStackIPO | Rhoda AI — Funding, Valuation & IPO Status | |
| SI017 | SEC | EDGAR search results for Rhoda AI | |
| SI018 | SEC | EDGAR search results for Rhoda Ai Corporation | No matching companies. |
| SI019 | California Companies Directory | Rhoda AI Corporation | California Companies Direcotry | |
| SI020 | Ashby | Rhoda AI Jobs | Open Positions (33) |
| SI021 | robotics.press | Rhoda AI: Competitive Response | robotics.press | Despite strong investor backing, the company lacks independently validated customers or disclosed revenue. |
| SI022 | The Robot Report | Rhoda AI exits stealth with $450M to train robots from video | |
| SI023 | Humanoids Daily | Rhoda AI Hits $1.7B Valuation, Unveils "Direct Video-Action" Model to Bridge the Real-World Gap | |
| SI024 | US Finance Insider | AI Robotics Startup Rhoda Hits US$1.7 Billion Valuation after Successful Funding Round | |
| SI025 | Ashby | VP of Hardware @ Rhoda AI | |
| SI026 | Ashby | Supply Chain & Logistics Lead @ Rhoda AI | |
| SI027 | Ashby | Inference Infrastructure Engineer @ Rhoda AI | |
| SE001 | Rhoda AI | Rhoda AI | FutureVision brings the capability to handle real world industrial tasks autonomously. |
| SE002 | Rhoda AI | Causal Video Models Are Data-Efficient Robot Policy Learners | Our Direct Video-Action Model (DVA) reformulates robot policies as video generation. |
| SE003 | Rhoda AI | Press Release | Rhoda AI | The resulting system continuously observes its environment, predicts future states as video, converts those predictions into actions, executes them, and re-observes the world. |
| SE004 | Rhoda AI | Team | Rhoda AI | |
| SE005 | BusinessWire | Rhoda AI Exits Stealth with $450 Million Series A to Bring Robots Out of the Lab and Into the Real World | |
| SE006 | RoboticsTomorrow | Rhoda AI Exits Stealth with $450 Million Series A to Bring Robots Out of the Lab and Into the Real World | |
| SE007 | Robotics and Automation News | Rhoda AI launches with $450 million Series A to bring robots out of the lab and into the real world | |
| SE008 | Humanoids Daily | Rhoda AI Hits $1.7B Valuation, Unveils "Direct Video-Action" Model to Bridge the Real-World Gap | |
| SE009 | RoboHorizon | Rhoda AI Unveils Video-Trained Robots, Nabs $450M at $1.7B Valuation | |
| SE010 | Interesting Engineering | Robot AI trained on millions of videos aims to work beyond labs | The company says FutureVision will eventually serve as a foundation model that can be licensed to partners building robotic hardware and software platforms. |
| SE011 | Coey | Rhoda AI’s Direct Video-Action Model Wants to Make Robots “Web-Trained” and Factory-Ready | The exact conditions vary by demo and are not yet standardized by third-party benchmarking. |
| SE012 | Assembly Magazine | New Robotic AI Platform Targets High-Variability Manufacturing Tasks | The system’s video-based pretraining allows it to learn new tasks quickly — often with as little as 10 hours of teleoperation data. |
| SE013 | CareersInRobotics | Rhoda ai Careers | 6 jobs | |
| SE014 | Mayfield | Cloud Infrastructure Engineer | Rhoda AI | |
| SE015 | Ashby | Fullstack Engineer @ Rhoda AI | |
| SE016 | YouTube | Rhoda AI: Returns Processing Demo | |
| SE017 | YouTube | Rhoda AI: Container Breakdown Demo | |
| SE018 | arXiv | GR00T N1: An Open Foundation Model for Generalist Humanoid Robots | |
| SE019 | arXiv | GR-2: A Generative Video-Language-Action Model with Web-Scale Knowledge for Robot Manipulation | |
| SE020 | arXiv | DreamGen: Unlocking Generalization in Robot Learning through Video World Models | |
| SE021 | arXiv | Foundation Model Driven Robotics: A Comprehensive Review | |
| SE022 | Kempner Institute | Large Video Planner: A New Foundation Model for General-Purpose Robots | |
| SE023 | Mimic Robotics | Video-Action Models: Are video model backbones the future of VLAs? | |
| SE024 | GitHub | Awesome Robot Foundation Models 2025-2026 | |
| SE025 | Google DeepMind | Introducing Gemini Robotics and Gemini Robotics-ER, AI models designed for robots to understand, act and react to the physical world. | |
| SE026 | JobScroller | Rhoda AI Jobs (June 2026) – 33 Open Roles | |
| SE027 | YouTube | Rhoda AI - YouTube | |
| SU001 | Rhoda AI | Rhoda AI homepage | We work with a variety of customers across verticals in automotive, manufacturing, logistics, and ecommerce. |
| SU002 | Rhoda AI | Rhoda AI Exits Stealth with $450 Million Series A to Bring Robots Out of the Lab and Into the Real World | The $450 million Series A will support continued research and engineering investment, expansion of industrial deployments and customer pilots. |
| SU003 | Business Wire | Rhoda AI Exits Stealth with $450 Million Series A to Bring Robots Out of the Lab and Into the Real World | |
| SU004 | RoboticsTomorrow | Rhoda AI Exits Stealth with $450 Million Series A to Bring Robots Out of the Lab and Into the Real World | |
| SU005 | TechNode Global | Temasek-backed Rhoda AI raises $450m Series A funding to accelerate robotics development | |
| SU006 | Robotics & Automation News | Rhoda AI launches with $450 million Series A to bring robots out of the lab and into the real world | |
| SU007 | Humanoids Daily | Rhoda AI Hits $1.7B Valuation, Unveils Direct Video-Action Model to Bridge the Real-World Gap | |
| SU008 | RoboHorizon | Rhoda AI Unveils Video-Trained Robots, Nabs $450M at $1.7B Valuation | |
| SU009 | AgentMarketCap | Rhoda AI $450M Series A: Video-Predictive Robotics and Physical AI Agents | |
| SU010 | Rhoda AI | Direct Video-Action Models | We present two example customer tasks, both of which were deployed as real customer proof of concepts and operated successfully for multiple hours without human intervention. |
| SU011 | Rhoda AI | Rhoda AI careers | |
| SU012 | Rhoda AI | Rhoda AI team | |
| SU013 | Reuters / Yahoo Finance | Rhoda AI raises $450 million, unveils platform for industrial environments | Industry experts caution that reliability, safety certification and cost will remain key hurdles for large-scale commercial deployment of general-purpose robots. |
| SU014 | Tech Funding News | Rhoda AI $450M Series A stealth exit robotics | |
| SU015 | Wilson Sonsini | Wilson Sonsini advises Rhoda AI on $450 million funding as company emerges from stealth | |
| SU016 | NIST | Robotic Systems for Smart Manufacturing Program | Yet, it is estimated that only 10% of potential users in the manufacturing domain have adopted robotic systems. |
| SU017 | Automate | How to Solve Robot Interoperability Issues in Industry 4.0 Manufacturing | |
| SU018 | MDPI Processes | Recent Advances and Challenges in Industrial Robotics: A Systematic Review | |
| SU019 | PwC | Global Supply Chain report | |
| SU020 | Mordor Intelligence | Warehouse Automation Market Analysis | The Warehouse Automation Market size is expected to increase from USD 29.98 billion in 2025 to USD 34.17 billion in 2026. |
| SU021 | MarketsandMarkets | Artificial Intelligence (AI) Robots Market | |
| SU022 | NVIDIA News | NVIDIA and global robotics leaders take physical AI to the real world | |
| SU023 | Monocle | Warehouse Automation ROI: Why 90% of Models Fail in 2026 | Only 10% of companies achieve sustained, large-scale success scaling automation beyond pilot programs. |
| SU024 | Rhoda AI / Ashby | Rhoda AI job board embed | |
| SU025 | RoboHorizon | Rhoda AI: broad industrial market intelligence layer | |
| SU026 | Amazon | Amazon hires from AI robotics startup Covariant, licenses technology | |
| SU027 | KNAPP | KNAPP and Covariant Extend Their Success Story | |
| SU028 | Modern Materials Handling | GXO pilots AI-enhanced robotics in warehouse | |
| SR001 | OSHA | Robotics | |
| SR002 | CDC / NIOSH | Center for Occupational Robotics Research | |
| SR003 | EUR-Lex | Regulation (EU) 2024/1689 Artificial Intelligence Act | |
| SR004 | European Commission | Machinery sector and legislation | |
| SR005 | ISO | ISO 10218-1:2011 Robots and robotic devices — Safety requirements for industrial robots — Part 1: Robots | |
| SR006 | ISO | ISO 10218-2:2011 Robots and robotic devices — Safety requirements for industrial robots — Part 2: Robot systems and integration | |
| SR007 | ISO | ISO/TS 15066:2016 Robots and robotic devices — Collaborative robots | |
| SR008 | Harvard Journal of Law & Technology | Redefining the Standard of Human Oversight for AI Negligence | |
| SR009 | Brookings Institution | Products liability law as a way to address AI harms | |
| SR010 | NIST | Robotic Systems for Smart Manufacturing Program | |
| SR011 | Automate | How to Solve Robot Interoperability Issues in Industry 4.0 Manufacturing | |
| SR012 | IndustrialEngineer.ai | The ROI-Driven Approach to Warehouse Robotics Integration | |
| SR013 | Cleverence | 3PL Robotics ROI Case Study: What Happened After the Investment | |
| SR014 | CNBC | Investors bet humanoid robots will transform industry and homes over the next decade | |
| SR015 | Humanoids Daily | Stealth startups emerge with over $300 million to join crowded humanoid robot field | |
| SR016 | Rhoda AI | Rhoda AI homepage | |
| SR017 | Rhoda AI | Rhoda AI press release | |
| SR018 | Rhoda AI | Direct Video-Action Models | |
| SR019 | Rhoda AI | Rhoda AI team | |
| SR020 | Rhoda AI | Rhoda AI careers | |
| SR021 | Robotics & Automation News | Rhoda AI launches with $450 million Series A to bring robots out of the lab and into the real world | |
| SR022 | Humanoids Daily | Rhoda AI Hits $1.7B Valuation, Unveils Direct Video-Action Model to Bridge the Real-World Gap | |
| SR023 | RoboHorizon | Rhoda AI Unveils Video-Trained Robots, Nabs $450M at $1.7B Valuation | |
| SR024 | Business Wire | Rhoda AI Exits Stealth with $450 Million Series A to Bring Robots Out of the Lab and Into the Real World | |
| SR025 | Reuters / Yahoo Finance | Rhoda AI raises $450 million, unveils platform for industrial environments | |
| SR026 | Tech Funding News | Rhoda AI $450M Series A stealth exit robotics | |
| SR027 | Wilson Sonsini | Wilson Sonsini advises Rhoda AI on $450 million funding as company emerges from stealth | |
| SR028 | AgentMarketCap | Rhoda AI $450M Series A: Video-Predictive Robotics and Physical AI Agents | |
| SR029 | Stanford HAI | The 2026 AI Index Report | |
| SR030 | RoboHorizon | Rhoda AI: hardware-agnostic play | |
| SV001 | Rhoda AI | News | Rhoda AI | Rhoda AI today announced its public launch after 18 months in stealth. |
| SV002 | Business Wire | Rhoda AI Exits Stealth with $450 Million Series A to Bring Robots Out of the Lab and Into the Real World | The $450 million Series A will support continued research and engineering investment, expansion of industrial deployments and customer pilots. |
| SV003 | TNGlobal | Temasek-backed Rhoda AI raises $450M Series A funding to accelerate robotics development | Temasek-backed robotics firm Rhoda AI has raised $450 million in Series A funding. |
| SV004 | Wilson Sonsini | Wilson Sonsini Advises Rhoda AI on $450 Million Funding as Company Emerges from Stealth | Rhoda AI ... announced it has emerged from stealth and raised $450 million in a Series A fundraising round. |
| SV005 | AgentMarketCap | Rhoda AI’s $450M Series A Signals the Physical AI Agent Boom | 27 physical AI startups collectively raised more than $6 billion in Q1 2026 alone. |
| SV006 | Humanoids Daily | Rhoda AI Hits $1.7B Valuation, Unveils Direct Video-Action Model to Bridge the Real-World Gap | Rhoda AI ... announced a massive $450 million Series B funding round ... at a $1.7 billion valuation. |
| SV007 | RoboHorizon | Rhoda AI Unveils Video-Trained Robots, Nabs $450M at $1.7B Valuation | The investment ... catapults the Palo Alto-based company to a hefty $1.7 billion valuation. |
| SV008 | TechCrunch | Robotics software maker Skild AI hits $14B valuation | The startup has raised a $1.4 billion Series C round that values it at more than $14 billion. |
| SV009 | Business Wire | Skild AI Raises $1.4B, Now Valued Over $14B | The company grew from zero to about $30M revenue in just a few months in 2025. |
| SV010 | TechCrunch | Physical Intelligence is reportedly in talks to raise $1B, again | Co-founder Lachy Groom told TechCrunch the company has no timeline for commercialization. |
| SV011 | TechCrunch | Figure reaches $39B valuation in latest funding round | Figure ... raised a Series C funding round that values it at $39 billion. |
| SV012 | Sacra | Figure AI valuation, funding & news | Figure AI reached a $39 billion post-money valuation in September 2025 following a Series C funding round that exceeded $1 billion in commitments. |
| SV013 | TechCrunch | Yet another AI robotics firm lands major funding, as Dexterity closes latest round | Dexterity ... has raised $95 million at a post-money valuation of $1.65 billion, per Bloomberg. |
| SV014 | TechCrunch | FieldAI raises $405M to build universal robot brains | FieldAI ... has raised $405 million across multiple previously undisclosed rounds to develop what it calls foundational embodied AI models. |
| SV015 | CNBC | Apptronik raises $520 million to beat Chinese humanoids, Tesla Optimus to market | Apptronik raises $520 million at $5 billion valuation for Apollo robot. |
| SV016 | The Robot Report | Physical Intelligence raises $600M to advance robot foundation models | Physical Intelligence ... has raised a total of $1.1 billion to date and is currently valued at about $5.6 billion, according to Bloomberg. |
| SV017 | The Robot Report | Skild AI grabs $300M to build foundation model for robotics | Skild AI ... announced that it has closed a $300 million Series A round. The funding brings its valuation to $1.5 billion. |
| SV018 | Securities and Exchange Commission | Symbotic Inc. Annual Report on Form 10-K — Fiscal Year Ended September 27, 2025 | We have approximately $22.5 billion of backlog as of September 27, 2025. |
| SV019 | Yahoo Finance | Symbotic Inc. (SYM) Stock Price, News, Quote & History | Valuation Measures as of 6/4/2026: Market Cap 6.03B, Price/Sales 2.27, Revenue (ttm) 2.52B. |
| SV020 | Yahoo Finance | Zebra Technologies Corporation (ZBRA) Stock Price, News, Quote & History | Valuation Measures as of 6/5/2026: Market Cap 11.06B, Price/Sales 2.10, Revenue (ttm) 5.58B. |
| SV021 | Yahoo Finance | Rockwell Automation, Inc. (ROK) Stock Price, News, Quote & History | Valuation Measures as of 6/5/2026: Market Cap 49.71B, Price/Sales 5.72, Revenue (ttm) 8.8B. |
| SV022 | Rockwell Automation | Financials - Annual Reports & Proxy | As a public company, Rockwell Automation is required to file registration statements, periodic reports, and other forms with the U.S. Securities and Exchange Commission. |
| SV023 | Zebra Technologies | Zebra Technologies Corporation - Financials | Zebra Technologies Corporation - Financials. |
| SV024 | Eilla AI Insights | The Complete Valuation Playbook for Robotics Businesses | Warehouse & intralogistics robotics (systems/RaaS) ~2.2x-5.0x ... industrial automation integrators roughly 0.9x-2.1x EV/Revenue. |
| SV025 | MarketsandMarkets | Artificial Intelligence (AI) Robots Market Report 2025 - 2030 | Long time to commercialize robots and high maintenance costs are among the most significant challenges confronting the AI robots market. |
| SV026 | Rhoda AI | Team | Rhoda AI | Our Leadership Team ... Jagdeep Singh ... Eric Chan ... Gordon Wetzstein. |
| SV027 | NVIDIA | From Warehouse to Wallet: New State of AI in Retail and CPG Survey Uncovers How AI Is Rewiring Supply Chains and Customer Experiences | 90% said they’d build on the success of current projects by increasing their AI budgets in 2026. |
| SV028 | UPS Supply Chain Solutions | 2026 Supply Chain Outlook | Software-defined warehouses integrating enterprise systems, robotics and real-time data. |
| SV029 | McKinsey & Company | Unlocking the industrial potential of robotics and automation | Automation will represent 30 percent or more of capital spending in the next five years for logistics and fulfillment players. |
| SV030 | U.S. Chamber of Commerce | Understanding America’s Labor Shortage: The Most Impacted Industries | As of April 2025, a gap persists, with 313,000 durable goods manufacturing job openings yet to be filled. |
| SV031 | Bureau of Labor Statistics | Hand Laborers and Material Movers | About 1,008,300 openings for hand laborers and material movers are projected each year, on average, over the decade. |