初创公司尽调
尽调报告 robotics / hardware Series A 2026-10-14

Rhoda AI

有顶尖资本加持的实体 AI 平台,技术潜力可观,但商业披露有限。

Rhoda AI 兼具突出的具身 AI 野心与顶级资本支持,但商业披露过少,尚不足以让人有把握地认可当前 $1.7B 估值。

覆盖范围与披露说明

截至 2026-10-14,商业牵引力、定价、董事会治理和生产安全证据仍然有限。

封面要素

投后估值 01
1700 USDm [CV001]
总部 03
Palo Alto, CA, USA [CO001]

公司概况

Rhoda AI 是一家私有披露的实体 AI 与机器人公司,围绕 Direct Video Action 模型打造 FutureVision 平台,服务工业操作场景。公司把软件授权与自研移动操作机器人硬件结合起来,在大型 Series A 财团支持下走出隐身期,但尚未公开披露具名付费生产客户、经审计收入或详细定价。

官网
www.rhoda.ai
创始人
Jagdeep Singh, Eric Ryan Chan, Gordon Wetzstein
创立地点
Palo Alto, California
总部
Palo Alto, CA, USA
产品
FutureVision 是面向工业机器人的 Direct Video Action 基础模型平台,并配套自研轮式移动操作机器人硬件,服务制造与物流流程。
客户
工业制造、物流和电商履约运营商
商业模式
软件授权、劳动力即服务部署,以及自研机器人硬件销售
阶段
Series A
融资情况
2026 年 3 月宣布完成 $450M Series A,投后估值 $1.7B
[CO001, CO002, CO003, CO004, CO005, CV001]

执行摘要

主要优势

  • 创始团队与科研背景出色,投资人阵容资本实力雄厚。
  • 独特的 Direct Video Action 架构,瞄准高难度工业操作任务。
  • 早期试点证据显示,系统在若干受限工作流中已具备实质性自主能力。

主要风险

  • 没有公开具名的付费生产客户、续约信息或经核实的车队经济性。
  • 定价、毛利率和单机成本结构仍未披露。
  • 生产安全、可靠性和治理披露仍不完整,尚不足以支撑投资判断。

未决问题

  • 具名付费客户、部署数量、续约情况及客户背调电话。
  • 定价模式、软件抽成率、毛利率,以及任何 ARR 或收入分母。
  • 董事会与治理结构的准确情况,以及法律注册日期与实际运营创立时间线之间是否存在不一致。
  • 独立的生产安全验证,以及长时间车队可靠性证据。

目录

Chapter 01

01公司概览

1.1 公司身份、成立背景与核心使命

Rhoda AI 是一家获风险资本支持的实体人工智能与机器人公司,总部位于加利福尼亚州帕洛阿尔托,处于生成式基础模型与工业自动化的交汇处。公司成立于 2024 年底,约 18 个月以保密隐身方式开发,直到 2026 年 3 月凭借 $450M Series A 融资公开亮相,并由此建立 $1.7B 投后估值。公司的核心使命,是用能跨商业和工业流程泛化的基础模型,解决机器人长期存在的适应性瓶颈。Rhoda AI 不再为每条轨迹单独编程,也不再让人远程操控机器数千小时,而是把 FutureVision 定位为通用智能层,覆盖高产量汽车装配线、离散制造工厂、第三方物流履约中心和电商供应链中的实体任务。[CO001, CO003, CO005, CO006, CO007]

KPI 快照表
指标已披露数值 / 状态观察日期证据置信度尽调缺口评估
投后估值$1.7B2026-03-10高机构财团和领投方披露均已确认
Series A 募资额$450M2026-03-10高Reuters 和财团披露均验证了单轮股权融资
年度经常性收入(ARR)未披露(null)2026-10-14低未披露经审计的收入运行率、软件 ARR 或客户合同金额
活跃付费客户未披露(null)2026-10-14低汽车和物流试点正在进行,但没有一个具名参考客户
生产部署试点阶段2026-10-14中已展示商业试验;连续商业产线密度尚未验证
员工总数未披露(null)2026-10-14低运营团队规模尚未公布;研究、硬件和数据整理岗位正在招聘

截至 2026-10-14,根据 Series A 披露和公司声明汇总的运营及财务指标;null 表示尚未发布的私人运营数据。

[CO001, CO002, CO005, CO006, CO011, CO012]

1.2 管理层与科学基因

Rhoda AI 由 CEO Jagdeep Singh 领导。Singh 是资深深科技企业家,曾创办并推动固态电池制造商 QuantumScape Corporation 上市。他擅长管理高资本开支、研究密集型工程周期,以及多年商业化过程。核心 AI 架构由首席科学官 Eric Ryan Chan 负责;他是专攻生成式 3D 建模和世界表征系统的计算机视觉研究者,曾在 WorldLabs 从事生成建模研究。科学指导由斯坦福大学副教授、斯坦福计算成像实验室负责人 Gordon Wetzstein 主导。这个领导班子兼具融资能力和前沿生成视觉经验,但公司对 Singh 的关键人物依赖明显:战略方向、资本组局和硬件战略执行都集中于他;正式董事会治理和委员会分工仍未披露。[CO003, CO004, CO013]

领导层与创始人表
领导者姓名公司职务职业背景职能领域与创始人—市场匹配度关键人物依赖评估
Jagdeep Singh联合创始人兼首席执行官连续创业的深科技企业家;QuantumScape 创始人及前 CEO(NYSE: QS)高管领导、大规模资本调动、深科技制造扩张高;核心愿景负责人、媒体发言人和融资支柱
Eric Ryan Chan首席科学官生成式 AI 与计算机视觉研究者;前 WorldLabs 研究科学家Direct Video Action 架构、世界模型、生成式预训练流程高;视频预测表征与后训练方法的主要架构师
Gordon Wetzstein科学领导 / 联合创始人Stanford University EE 副教授;Stanford Computational Imaging Lab 负责人计算成像、神经渲染、视觉传感器处理、感知算法中;提供专门科学顾问支持,并负责前沿计算机视觉对齐

高管领导层和科学创始人在结束隐身运营的公告中披露;董事席位和治理名单仍属私人信息。

[CO003, CO004, CO013]

1.3 资本结构、估值基准与投资财团

Rhoda AI 的 $450M Series A 是实体 AI 与机器人领域规模最大的首轮风险融资之一,公司走出隐身期即成为独角兽。融资由 Wipro 创始人 Azim Premji 的投资机构 Premji Invest 领投,体现多阶段全球机构资本对实体自动化的激进配置。财团还包括 Khosla Ventures、Mayfield、Temasek Holdings 及其子公司 Xora Innovation、Capricorn Investment Group、Leitmotif、Matter Venture Partners、Prelude Ventures,以及 John Doerr 等天使投资人。投资方把 Rhoda AI 定位为面向全球体力劳动经济的 Labor as a Service 平台。然而,Rhoda AI 尚未披露经审计收入跑速、ARR 或毛利率基准,因此 $1.7B 估值与近期财务指标完全脱钩。[CO001, CO002, CO010, CO012]

利益相关方或投资者图谱
利益相关方实体投资者分类战略意义与财团角色关键尽调问题
Premji Invest领投机构投资者$450M Series A 领投方;提供多阶段资产负债表支持确认董事席位分配、投票协议和清算优先权结构
Khosla Ventures前沿 AI 风险投资机构早期深科技先行者;为基础 AI 模型提供战略背书审查后续融资轮的按比例认购权和参与门槛
Temasek Holdings / Xora Innovation主权财富基金与深科技部门覆盖亚太物流的全球工业触达和耐心资本平台查验国际部署渠道和地缘政治共同投资条款
Mayfield风险投资合作方早期倡导者,将 Rhoda 定位为面向体力劳动的 Labor as a Service 平台评估财团治理参与度和企业客户引荐
Capricorn Investment Group可持续技术投资者重视深科技自动化和先进制造的耐心资本评估投资期限是否与多年期机器人硬件开发相匹配
John Doerr个人天使投资者传奇硅谷风险投资领袖;提供网络背书和战略资源澄清个人投资结构和非正式顾问参与情况

财团参与方已从 Series A 融资披露中核实;持股比例和董事会构成未披露。

[CO001, CO002, CO010]

1.4 平台架构里程碑与商业部署现实

Rhoda AI 的技术主张以 FutureVision 为核心,其 Direct Video Action(DVA)架构把机器人控制改写为连续视频预测任务。模型先用数亿条公开互联网视频预训练,学习运动、物理和物体动力学的通用表征,再用 1 至 10 小时机器人专属轨迹数据进行后训练。闭环控制每几百毫秒运行一次,观察环境状态并预测电机动作,以应对布局变化和非结构化物体。在工业试验中,Rhoda 展示了每个周期不到两分钟完成自主部件处理。为商业化这一技术栈,公司把配备 7-DoF 机械臂和灵巧手的自研轮式移动操作机器人,与第三方软件授权结合起来。不过,关键尽调风险仍在:独立分析者指出,隐身期演示很少代表生产环境的波动;高 GPU 算力开销给边缘部署带来挑战;Rhoda 也尚未披露任何已签约、付费的商业客户。[CO007, CO008, CO009, CO011, CO014, CO015]

FO001: 公司里程碑时间线

Rhoda AI 从成立、隐身开发到平台发布、Series A 融资及商业扩张的里程碑时间序列。

里程碑日期源自官方 Series A 公告、Reuters 报道和高管访谈。

[CO001, CO003, CO004, CO007, CO008, CO009, CO014]

1.5 展品

Chapter 02

02市场分析

2.1 市场定义与范围:从刚性自动化向外扩展

实体人工智能的出现,意味着工业机器人从刚性预编程转向适应性、感知驱动的非结构化操作。传统汽车装配和仓储物流设施已在点焊、激光切割、标准托盘搬运等可预测重复流程中实现较高固定自动化渗透;但部件拆包、打开可变形容器、零件配套和包裹分拣等环境变化大的任务,长期难以自动化,仍依赖人工。Rhoda AI 瞄准的正是这条高变异操作边界。公司成立于 2024 年底,2026 年 3 月在 Premji Invest 领投的 $450M 融资支持下走出隐身期,试图在汽车和物流设施替代既有体力劳动。它不与成熟 PLC 或固定机械臂供应商正面竞争,而是聚焦物体朝向和动态布局不断变化、令传统 VLA 模型及远程操作轨迹系统失效的复杂实体交互任务。Rhoda AI 用数亿条互联网视频训练自研 DVA 架构,试图赋予机器人通用实体常识,把工业运营中可自动化的边界推进到过去被认为难以由机器完成的任务。[CM001, CM002, CM005, CM011, CM012]

物理 AI 操作的市场定义与范围
细分市场 / 类别纳入的支出排除的支出买方 / 付款方对 Rhoda AI 的相关性
汽车与重型制造可变卸料、高力容器拆解、零件套装、产线补料固定工装、机器人焊接单元、刚性预编程龙门输送机制造副总裁 / 工厂运营经理核心上市重点;已在真实工厂展示低于 2 分钟的循环时间
仓储与物流履约高可变料箱分拣、包裹卸料、退货处理、托盘卸货自动存取系统(ASRS)、固定分拣输送机、AGV 平面搬运首席供应链官 / 配送副总裁目标扩张市场;枯燥、肮脏、重复性人工岗位的人员流动率高
移动操作硬件集成式 7-DoF 机械臂、灵巧三指手、轮式移动底座、工业传感器固定式协作机器人、双足人形机器人腿部、消费级服务机器人自动化工程总监 / 资本设备采购负责人专有全栈系统,设计用于搬运 45kg 载荷并保持稳定急停
物理 AI 基础软件Direct Video Action 基础模型授权、边缘推理运行时、任务后训练传统 PLC 轨迹编程、标准计算机视觉目标检测 API企业自动化团队 / 首席技术官FutureVision 软件平台授权给第三方机器人 OEM

市场边界按任务可变性划定:Rhoda AI 瞄准难以用固定自动化解决的动态任务,同时排除刚性预编程输送机和焊接基础设施。

[CM001, CM002, CM005, CM006, CM007, CM011, CM012]

2.2 可服务市场测算:宏观劳动力套利与实际 SAM

风险投资人以巨大的宏观可服务市场评估 Rhoda AI,把机会描述为 Labor as a Service(LaaS),而非资本设备销售。Mayfield Fund 将 TAM 定义为全球体力劳动经济,覆盖制造、仓储、建筑和农业,每年人工工资支出约 $30T。按这一投资逻辑,实体 AI 不只是替代既有机器人预算,还能用可扩展算力和机械系统替代稀缺体力劳动,释放受限的工业 GDP。Mayfield 的经济模型假设,部署 1,000 台自主移动操作机器人并签订持续劳务服务协议,可产生约 $100M ARR;规模扩大到 10,000 台,则可形成 $1B ARR 企业。但要把理论上的劳动力套利变成 SAM,必须严格收窄边界。工业操作的近期可服务市场限于结构化室内制造工厂和高产量物流配送中心,且这些场所面临严重用工短缺和较高重复性劳损率。尽管 2026 年 1 月单月更广义的机器人与 AI 风险投资已超过 $4.3B,并推动 Rhoda AI 达到 $1.7B 估值基准,商业投资判断仍应在商业车队可靠性和多设施单位经济性经生产验证前,对自上而下的劳动力 TAM 打折。[CM003, CM004, CM008, CM010, CM011, CM013]

TAM/SAM/SOM 估值视角与经济基准
估值视角 / 估算发布方 / 来源年份地理范围数值CAGR / 增长方法论置信度局限 / 备注
全球体力劳动市场(TAM)Mayfield Fund / World Bank 劳动力估算2026全球$30.0 Trillion3-4% 年度工资增长汇总所有行业的全球体力劳动和手工劳动支出低抱负性宏观经济池;绝大多数是非结构化户外工作或无法自动化的工作
工业与物流机器人行业SuperbCrew / 行业报告2026全球$20.0+ Billion25-30% 年度增长机器人硬件和智能自动化的年度资本投入中汇总固定机器人、AGV 和新兴 AI 平台,而非纯基础模型
Rhoda 部署 1,000 台设备的规模(SOM 代理)Mayfield Fund(LaaS 预测)2026北美与欧洲$100.0 Million ARRN/A(里程碑)假设每台已部署机器人每年经常性合同金额约为 ~$100k 的单位经济性中公开客户合同披露或已签采购订单尚未验证该模型假设
Rhoda 部署 10,000 台设备的规模(SAM 代理)Mayfield Fund(LaaS 预测)2026全球工业中心$1.0 Billion ARRN/A(里程碑)假设以 $100k/年 LaaS 费率运营 10,000 台移动操作机器人低假设成功扩展到多工厂企业、硬件可靠,并且不会被竞争者取代

数值反映宏观经济规模估算和投资者投资判断预测;Rhoda AI 尚未公开报告当前 ARR、设备部署量或已实现合同金额。

[CM001, CM003, CM004, CM008, CM009, CM013]
FM001: Physical AI 的受限市场规模分层

从 $30T 全球体力劳动上限,到近期可服务可获取市场和初始试点部署里程碑的结构化拆解。

各层宽度仅表示层级,不代表数值比例。

宏观 TAM 代表全球体力劳动支出;中间层代表工业自动化支出和投资者建模的经常性收入目标。

[CM003, CM004, CM008, CM010, CM011]

2.3 买方细分:工厂经济性、安全约束与形态偏好

Rhoda AI 目标行业的企业采购由工厂运营经理、职场安全负责人和资本配置委员会主导,回本门槛严格。汽车制造和物流配送中心的典型买方要求新自动化系统在 12 至 24 个月内完成资本回收,衡量标准是替代人工、提升吞吐并降低人体工学风险。学术机器人演示与商业客户对形态的偏好之间出现明显分歧:风险资本大量补贴试图复制人体结构的双足人形机器人,但工业运营高管更偏爱轮式移动操作机器人。轮式底盘可以与 AGV、AMR 一起融入现有工厂交通布局,避开双足行走固有的严重倾倒风险。公司管理层和行业协会观察指出,双足人形机器人急停可能发生灾难性倒地,损坏周边设备或伤及人员;轮式操作机器人则能保持静态稳定。Rhoda AI 选择轮式底盘、七自由度机械臂、三指灵巧手和 45 kg 负载能力,正是为了贴合工业工厂的安全标准与采购预期。[CM006, CM007, CM010, CM011, CM012]

细分市场与买方决策矩阵
市场细分目标买方最终用户付款方 / 预算负责人目标工作流采用触发因素回本期 / ROI 门槛
汽车 Tier-1 供应商与 OEM制造运营副总裁工厂装配技师和产线主管工厂资本支出 / 运营预算零部件卸料(10kg 零件)、容器拆解、套装产线操作员严重短缺、人体工学伤害、重复性劳损12-18 个月回本;必须超过人工循环速度(低于 2 分钟)
电商物流与履约配送副总裁 / 全球物流负责人仓库履约员工和分拣员运营与自动化 OPEX 预算退货处理、拆箱、包裹分拣、料箱转运旺季劳动力短缺、仓库年度流动率超过 100%基于完全负担人工替代和吞吐量,回本期低于 12 个月
工业包装与仓储供应链自动化总监物料搬运团队和叉车操作员工厂维护与现代化预算清空 50-pound Contico 容器、托盘拆垛安全合规事故、重物搬运的身体限制18-24 个月回本;要求与现有 AMR/AGV 流量无缝集成
第三方机器人 OEM 与集成商产品副总裁 / 机器人工程负责人OEM 应用工程师和软件开发者研发 / 技术授权预算将 FutureVision 软件集成到合作方机械臂和移动底座无法自行构建物理世界模型和闭环 DVA 流程相比招聘 50+ 名专门的物理 AI 研究人员,具备授权利润率优势

买方画像综合工业合作伙伴反馈、行业协会调查和高管访谈;试点转化率尚未披露。

[CM002, CM005, CM006, CM007, CM010, CM011, CM012]

2.4 增长驱动、商业障碍与估值尽调缺口

实体 AI 平台的采用速度受到宏观顺风推动,也被严格的运营和财务瓶颈限制。主要驱动包括制造业长期缺工、工业劳动力老龄化,以及北美和欧洲企业回流带来的非线性生产率需求。Rhoda AI 在一家真实汽车工厂完成每周期不到两分钟的自主部件处理流程,初步试点说明视频预测闭环控制可以满足企业周期时间要求。但重要的采用约束仍未解决:工业买方要求六西格玛可靠性(99.9999% 正常运行时间)、共享人机工作单元的稳健安全认证,以及无需工程师介入的快速边缘故障恢复。Rhoda AI 的商业轨迹还被重大尽调缺口遮蔽:公司未公开合同客户数、部署数量、试点转化率或软件授权毛利率。$1.7B 投后估值和每股 $35.17 的股权定价体现的是风险资本乐观情绪,而非已建立的经常性现金流。要跨过试点技术可行性与持久商业采用之间的鸿沟,Rhoda AI 必须证明 DVA 模型能在不同制造工厂保持运营可用性,且无需昂贵的定制后训练或专门远程操作监督。[CM004, CM005, CM006, CM009, CM010, CM014]

2.5 展品

Chapter 03

03竞争对手

3.1 竞争格局结构与战略类型

Rhoda AI 所处竞争格局分为四类:纯机器人基础模型开发商、垂直整合的人形机器人 OEM、既有工业自动化厂商,以及企业内部自动化建设。第一类包括 Skild AI 和 Physical Intelligence(Pi)等实体 AI 基础模型开发商,侧重模型架构和硬件无关授权。Skild AI 于 2026 年 1 月以 $14B 估值完成 $1.4B Series C;Physical Intelligence 也专注跨多种机身的通用机器人模型。Rhoda 与纯软件供应商不同,在开发 FutureVision 的同时自研移动操作机器人硬件,而非立即提供通用模型授权。第二类包括 Figure AI 和 Apptronik 等垂直整合人形机器人制造商:Figure AI 到 2026 年初估值已达 $39B、融资超过 $1B;Apptronik 于 2026 年 2 月获得 $520M Series A 延伸融资,估值超过 $5B、累计融资 $935M。与存在动态平衡故障模式的双足架构相比,Rhoda 采用轮式底盘、七自由度机械臂和三指灵巧手的移动操作机器人,提供 45 kg 负载能力,并在急停时消除倾倒风险。第三类是 FANUC、ABB、KUKA 等传统工业机器人巨头,以确定性编程和成熟全球现场支持主导固定工作单元;第四类则包括一级制造商和物流运营商内部的机器人工程团队。[CP001, CP002, CP003, CP004, CP005, CP006, CP007]

竞争者画像表
竞争者 / 替代方案类别规模 / 融资目标细分市场核心差异化战略局限
Rhoda AI混合基础模型与 OEM$450M Series A($1.7B 投后估值)汽车装配、制造、物流退货Direct Video-Action 视频预测控制、轮式移动操作机器人企业试点受限;公开 API 文档和 SDK 尚未发布
Skild AI机器人基础模型纯玩家$1.4B Series C($14B 估值,2026 年 1 月)跨行业机器人 OEM、操作、运动与硬件无关的基础智能平台,资本规模庞大依赖第三方硬件集成和 OEM 数据共享条款
Figure AI垂直整合人形机器人 OEM$39B 估值,截至 2026 年初融资超过 $1B汽车制造、商业物流全栈双足人形形态、知名企业合作双足动态平衡风险、单位资本成本高、现场寿命尚未验证
Apptronik人形机器人 OEM 平台$520M Series A 增资,估值超过 $5B,累计融资 $935M工业制造、物流、供应链模块化商用人形机器人、Apollo 平台开发硬件开发负担重,商业部署仍处早期
Physical Intelligence(Pi)机器人基础模型纯玩家获顶级 AI 投资者风险投资支持通用机器人操作、商业设施多本体策略学习,专注通用物理智能没有用于受控物理数据采集的自有工业硬件机队
工业在位者(FANUC / ABB / KUKA)现状工业自动化数十亿美元规模的上市企业结构化汽车、电子、重型制造企业分销深厚,结构化单元可靠性超过 99.9%在非结构化环境中脆弱,缺乏生成式自适应智能

规模和融资指标反映截至 2026 年初披露的融资轮次和估值。私有基础模型定价和单位经济性仍未披露。

[CP001, CP002, CP003, CP004, CP005, CP007, CP013]
FP001: 竞争定位图

战略定位聚焦硬件集成广度与模型泛化范式。

条目X (-5–4)Y (-4–5)说明
1. Rhoda AI 4 3 Video-predictive DVA foundation model paired with proprietary wheeled mobile manipulator
2. Skild AI 4 -4 Generalist robot foundation model pursuing hardware-agnostic cross-embodiment licensing
3. Figure AI 2 5 Vertically integrated full-stack bipedal humanoid robotics with massive venture capital backing
4. Apptronik 1 4 Turnkey commercial humanoid developer focused on supply chain and manufacturing tasks
5. Physical Intelligence 3 -4 Pure-play physical AI foundation model developer without proprietary robot hardware
6. FANUC / ABB / KUKA -5 4 Traditional industrial robotics incumbents with deterministic programming and dedicated hardware

坐标轴代表基于公司披露材料中的架构范围和硬件耦合程度得出的定性尽调评分。

[CP001, CP005, CP006, CP007, CP011, CP014]

3.2 竞争者画像、技术架构与能力对比

竞争者能力揭示了 VLA 模型、重度远程操作流程和 Rhoda DVA 架构之间的根本差异。主流 VLA 处理图像 token 和文本提示来推断动作序列,但独立分析和行业评论认为,主流 VLA 通常只覆盖很短的上下文窗口,只有几帧视频。这会削弱其在长时程跟踪或连续物体操作中的准确率。传统机器人策略训练还高度依赖昂贵的硬件远程操作,掌握新任务往往要数千小时机器人专属演示。相比之下,Rhoda 先用数亿条互联网视频预训练 FutureVision,建立实体先验,再用约 10 小时机身专属远程操作数据微调,闭环控制每几百毫秒更新一次。不过,Skild AI 等纯基础模型竞争者拥有更大的资本储备($1.4B 轮次)和广泛的硬件无关、多机身合作;Figure AI 等人形机器人开发商则利用自有双足形态,直接瞄准人形物流与制造流程。传统工业机器人厂商在结构化生产线上保持近 100% 的确定性可靠性,而基础模型平台仍处于运营试点验证阶段。[CP002, CP004, CP005, CP008, CP009, CP010, CP011, CP012]

功能 / 能力矩阵
能力维度Rhoda AISkild AIFigure AI传统工业机器人证据基础 / 尽调备注
架构家族Direct Video-Action(DVA)闭环与机器人无关的基础模型VLA / 多模态端到端策略确定性轨迹脚本 / PLC主要新闻稿、技术分析和公司披露
预训练数据基础数亿条互联网视频跨本体仿真与真实数据多模态网络数据与遥操作无预训练;编程设定的工作单元几何Rhoda 新闻稿和行业报道引用了网络视频规模
新任务适配时间约 10 小时遥操作数据少样本跨本体迁移数小时遥操作与微调数天至数周手工 CAD / 离线编程Rhoda 声称适配约需 10 小时,而行业标准需数千小时
控制频率 / 循环每几百毫秒进行一次闭环预测高频策略执行连续实时策略控制毫秒级确定性控制回路Rhoda 多份技术概览均确认闭环视频预测
视觉上下文记忆长上下文视觉历史(数百帧)多步骤视觉记忆通常只有几帧的视觉缓存没有时间记忆的静态相机校准独立技术分析强调 DVA 的视觉记忆优于 VLA 的限制
硬件形态轮式移动操作机器人(7-DoF 机械臂)与硬件无关(授权给第三方)完整双足人形形态固定式关节机器人臂 / 龙门架CEO 访谈确认轮式底座配备 7-DoF 机械臂和 45kg 载荷

能力对比涵盖同类公司类别中的架构主张和已验证设计模式。生产环境中的具体自主成功率仍受客户 NDA 约束。

[CP005, CP006, CP008, CP009, CP010, CP011]

3.3 商业包装、切换成本与长期护城河

实体 AI 的商业包装仍在形成,竞争者在硬件销售、Robotics-as-a-Service(RaaS)和基础模型授权之间采用不同变现机制。成熟工业机器人厂商以资本设备销售为主,前期硬件成本搭配长期维护合同和系统集成服务,由固定工厂工具及专有控制器语言带来高切换成本。Figure AI、Apptronik 等人形机器人厂商正在探索面向商业企业设施的交钥匙软硬件租赁或 RaaS 订阅。对基础模型开发商而言,软件授权是主要模式,但 Skild AI 和 Physical Intelligence 等纯平台尚未公布标准化公开 API 定价。Rhoda 的计划是双重变现:先部署自研移动操作机器人,用于数据收集和验证;随后把 FutureVision 作为智能层授权给第三方机器人硬件与软件厂商。然而,独立行业评测指出,Rhoda 上线时缺乏公开 API 文档、SDK 和作业控制端点,更像企业试点阶段,而不是可立即调用的平台。其护城河能否持久,取决于网页视频预训练和 DVA 闭环架构能否在资金充足的基础模型竞争者或人形 OEM 占满企业试点名额前,启动更优的真实世界数据飞轮。[CP002, CP003, CP007, CP012, CP013, CP014, CP015, CP016]

定价 / 产品打包对比
竞争者 / 模型包装 / 交付机制硬件所有权定价披露状态切换成本与锁定动态
Rhoda AI直接部署移动操作机器人;计划通过 FutureVision 授权专有轮式硬件,另加合作伙伴授权私有 / 未披露(企业试点条款)中等:硬件集成和工作流定制会造成客户惯性
Skild AI向机器人 OEM 和企业开发者授权基础模型无(硬件无关)私有 / 未披露(定制商业授权)一旦嵌入 OEM 固件,软件集成带来较高切换成本
Figure AI交钥匙人形机器人租赁、RaaS 及企业试点合同专有全栈双足人形机器人私有 / 未披露(企业战略合同)设施定制集成和机群编排带来较高切换成本
Apptronik商用人形平台硬件销售及企业试点专有模块化人形硬件(Apollo)私有 / 未披露(商业试点条款)资本投入高,且需要专门的运营培训
Traditional Automation (FANUC/ABB)前期设备采购 + 系统集成 + SLA 维护专有工业关节臂及控制器硬件公开目录价($30k-$150k+),另加定制集成极高:专有工具、固定生产线安全围栏和 PLC 基础设施

早期基础模型初创公司的目录价完全未披露;企业合同反映定制化商业试点和战略合作里程碑。

[CP001, CP002, CP003, CP012, CP014, CP015]

3.4 展品

Chapter 04

04财务

4.1 收入来源、授权策略与定价架构

Rhoda AI 正在搭建以实体智能基础能力为核心的商业架构,面向制造、汽车拆包和履约物流中的工业机器人。核心商业化引擎是 FutureVision:由 DVA 预测模型驱动的基础模型平台,计划授权给第三方机器人 OEM 和自动化软件平台。同时,公司开发配备七自由度机械臂、三指灵巧手和轮式底盘的自研移动操作机器人,可持续处理 25 kg、峰值处理 40 kg 负载。由此形成双重变现:高毛利的经常性软件授权,配合资本密集型全栈机器人部署。领投方 Mayfield 还提出 Labor-as-a-Service(LaaS)框架,客户为自主完成的工作付费而不是购买设备;按其模型,1,000 台部署机器人可产生约 $100M ARR,10,000 台则达到 $1B ARR。但 Rhoda 目前完全没有收入,未披露商业合同、经常性软件订阅或公开标价,所有未来收入预测都只是未经验证的假设。[CI003, CI004, CI005, CI006, CI007, CI008, CI013]

Rhoda AI 商业收入来源与变现机制
收入来源变现机制单位 / 合同指标当前商业状态收入质量画像主要尽调问题
FutureVision 基础模型授权向机器人 OEM 收取平台软件授权费和运行时推理版税按机群或活跃机器人设备收取年度授权费部署前;架构于 2026 年 3 月公布高毛利经常性软件收入(毛利率潜力约 75-85%)OEM 合作协议、版税计划和最低承诺条款
专有移动操作机器人销售销售集成 DVA 智能的全栈轮式机器人硬件按物理设备收取资本开支采购价内部开发及早期工业客户测试较低毛利的硬件销售(毛利率约 30-40%),会占用营运资金硬件物料清单、制造成本曲线和保修准备金
劳动力即服务(LaaS)部署提供全托管自动化服务,按完成的生产性工作收费年度订阅费(目标约 $100k/台)或按小时计费Mayfield 主导投资人建模的探索性商业框架经常性收入可预测,但持续产生服务、维护和机群保养成本SLA 承诺、单机维护成本和客户流失假设
工业试点与定制服务收取非经常性工程(NRE)及工作流集成概念验证费按生产设施或工作流单元收取固定里程碑费用正在进行生产评估(例如,组件处理周期 <2 min)服务收入毛利低且非经常性,但对试点转化至关重要试点转商业合同的转化率及客户成本承担情况

收入来源反映已披露的商业架构和投资人框架;商业合同和实际定价仍需通过一手尽调核实。

[CI001, CI003, CI004, CI006]
定价架构与变现基准分析
商业层级 / 产品定价模型与合同基础目录价与实际价格状态折扣与商业未知项证据来源 / 基准依据
劳动力即服务部署隐含每台部署设备约 $100,000 的年度经常性费用投资人经济模型(1,000 台对应 $100M ARR);未公开列价数量折扣曲线、多年锁定条款和正常运行时间罚金未披露Mayfield 投资备忘录
FutureVision OEM 软件授权按机器人收取分级年度运行时费用,另加按任务收取的训练费商业定价表未披露;正在与合作伙伴讨论OEM 集成补贴、最低机群承诺和收入分成比例未知Rhoda AI 发布公告
交钥匙工业试点项目预计每个设施评估工作流 $150,000 - $350,000私有商业安排;试点条款未披露客户分摊成本与风险投资补贴评估条款尚未核实Automate.org CEO 访谈
专有硬件采购选项预计前期硬件 CAPEX 为 $80,000 - $150,000,另加软件订阅费硬件价格尚未公布;正在与客户测试保修范围、硬件更换周期和残值未披露SuperbCrew 机器人行业报道

单台隐含费率反映风险投资建模;官方商业目录价仍属机密,需要通过客户审计核实。

[CI003, CI005, CI007, CI013]
FI001: Labor-as-a-Service 单位经济贡献示意瀑布图

假设的年度单位经济价值桥,展示客户劳动力替代价值与估算机器人直接运营成本的对比。

条目数值原文标注说明
人工替代总价值 140000 USD/year 每个工作单元替代 2 个手工班次后的基准完全负担成本 direction: increase direction: increase USD/year 每个工作单元替代 2 个手工班次后的基准完全负担成本
客户采用激励折扣 -40000 USD/year 让利给工业客户、用于证明切换自动化合理性的经济利润 direction: decrease direction: decrease USD/year 让利给工业客户、用于证明切换自动化合理性的经济利润
Labor-as-a-Service 净实现收入 100000 USD/year 按投资者模型,每台部署设备目标年度经常性收入为 $100K direction: total direction: total USD/year 按投资者模型,每台部署设备目标年度经常性收入为 $100K
硬件摊销与现场维护 -25000 USD/year 移动操作机器人硬件折旧成本及常规机械维护 direction: decrease direction: decrease USD/year 移动操作机器人硬件折旧成本及常规机械维护
模型推理与云计算 COGS -15000 USD/year 持续视频预测控制的计算开支和连接成本 direction: decrease direction: decrease USD/year 持续视频预测控制的计算开支和连接成本
估算单位贡献利润率 60000 USD/year 预计每台机器人毛贡献利润率约为 60%,尚未扣除中央研发分摊 direction: total direction: total USD/year 预计每台机器人毛贡献利润率约为 60%,尚未扣除中央研发分摊

瀑布图数值是基于 Mayfield 风投评论和工业自动化成本基准的示意模型;实际单位成本仍未公开。

[CI007, CI008, CI014]

4.2 成本结构、毛利率动态与硬件交付驱动因素

Rhoda 实体 AI 模式的财务可行性高度取决于单位成本结构和 DVA 的计算强度。不同于每项操作技能需要数百小时远程操作数据的模仿学习平台,Rhoda 用互联网规模视频预训练来建立物理先验,把后训练需求降至 1 至 10 小时机器人专属轨迹数据,从而显著提升数据效率、压低技能接入成本。但运营交付会带来明显 COGS 压力:闭环视频预测控制每秒运行数十次,需要专用高性能机载边缘加速器或持续的低延迟云推理,都会产生显著电力、散热或带宽成本。自研轮式移动操作机器人的制造和维护还会产生硬件折旧、执行器更换和现场维护费用,而传统软件供应商无需承担这些成本。CEO Jagdeep Singh 指出,现有工业人形机器人原型是无法持续完成工业班次的脆弱玩具,说明打造坚固、生产级硬件需要大量持续维护储备。软件授权毛利率约 75%–85%,全栈硬件交付约 35%–50%,两者差异很大,因此必须严格评估混合公司毛利。[CI009, CI010, CI014, CI015]

单位经济与财务投资判断指标
单位经济指标已披露 / 估算值投资判断置信度战略财务重要性必要尽调问题
每台部署设备的年度经常性收入$100,000(隐含模型)中决定支撑 $1.7B 投后估值所需的机群规模合同中的 SLA 定价条款及客户支付意愿数据
硬件物料清单(BOM)成本低划定全栈机器人硬件交付的毛利率上限执行器、传感器、电池及计算组件采购成本
每周期模型服务与边缘计算成本低Direct Video Action 要求每几百毫秒持续进行闭环推理云端推理与机器人端边缘芯片的运营开支
远程操作任务训练时长1 to 10 hours高决定部署启动成本和客户实现价值的时间不同工业工作流中的实证远程操作工时
客户回本周期低决定企业工业自动化采购速度的关键指标制造业试点评估中的 ROI 计算模型

null 值代表未在公开披露中公开的私有运营数据;需要在数据室中进行实证核实。

[CI007, CI008, CI009, CI010]

4.3 资本充足性、现金消耗轨迹与融资依赖

Rhoda 的财务跑道由 2026 年 3 月完成的 $450M Series A 支撑,这是工业机器人史上规模最大的早期融资事件之一。Khosla Ventures、Temasek、Mayfield、Premji Invest 和 Capricorn Investment Group 等顶级机构为公司提供了可观流动性。管理层计划把资本投入三条重资产路径:扩大模型预训练算力集群、与制造合作伙伴扩展生产试点部署,以及把技术团队从最初的 100 多人继续做大。尽管资产负债表充裕,未来现金消耗预计仍很高。用数亿条网络视频训练生成式视频模型需要数千万美元算力基础设施;硬件工程、测试设施和客户试点补贴也会快速消耗营运资金。在商业化仍处于探索性试点阶段时,Rhoda 在生产试点转为能产生现金的企业合同前,完全依赖风险资本和战略资产负债表支持。当前储备预计可支撑 24 至 36 个月运营,但硬件生产扩张还需要在现金流盈亏平衡前获得大量额外融资。[CI001, CI002, CI010, CI013]

4.4 财务结论、估值审视与投资判断尽调障碍

从机构投资判断角度看,Rhoda AI 具有双重面貌:技术野心和资本支持极强,但可验证商业牵引力完全缺失。没有确认客户收入或合同积压,却凭 Series A 达到 $1.7B 投后估值,必然面临市场密切审视。要与 Boston Dynamics 等成熟机器人开发商,以及 Tesla Optimus 等汽车厂商支持的人形项目竞争,公司必须迅速在工厂地面证明切实投资回报。行业分析者强调,工业企业投入数百万美元部署前,必须先跨过包括超过 99.9% 运营可靠性、共享人机工作单元复杂安全认证和总拥有成本在内的重大障碍。对投资人而言,主要尽调阻碍仍未缓解:Rhoda 未披露客户试点商业条款、初评转化率、软件定价表或单位硬件物料清单成本。在管理层提供经审计的试点转化经济性和已签客户承诺前,投资人必须把当前企业价值视为由风险资本驱动的知识产权资本化,而非有商业基本面支撑的业务。[CI002, CI011, CI012, CI013]

4.5 展品

Chapter 05

05产品与技术

5.1 产品定义、基础智能与硬件形态

Rhoda AI 打造全栈机器人智能平台,把实体机器人控制定义为实时因果视频预测,而不是传统轨迹编码或离散动作 token 策略。核心软件 FutureVision 面向工业物流和装配设施中的动态现实变化,材料配置和工厂布局持续变化也能应对。不同于预编程机械臂或专用固定功能自动化,Rhoda AI 把基础模型与面向商业耐久性的自研移动操作机器人结合起来。实体平台配备全向轮式底盘、两条七自由度机械臂和三指灵巧手,可持续处理 25 kg、靠近底盘时处理最多 45 kg 负载。公司放弃双足人形腿,改用天然稳定的轮式底盘,从而消除紧急断电时的灾难性倾倒和跌落风险,直接兼容现有 AGV 和仓储车队标准。[CE001, CE005, CE009, CE010, CE011]

产品模块与硬件平台矩阵
模块 / 平台组件目标用户 / 操作者技术成熟度核心差异化尽调信息缺口
FutureVision 智能平台物流与仓储工程师试点部署具备长上下文记忆的视频预测控制OEM 授权条款和 API 延迟基准未发布
Direct Video-Action(DVA)模型机器人 AI 研究人员与集成商研究原型 / 试点以 1M+ 网络视频预训练;上下文摊销视频模型与动力学模型之间的梯度流已解耦
移动操作机器人底座制造线主管客户测试轮式稳定性;急停时零倾覆风险多班次使用下的现场 MTBF 和维护生命周期
灵巧操作臂与手装配线操作员客户测试7-DOF 机械臂配三指手,可举升 25-45 kg磨蚀性负载下的夹具磨损率和传感器退化

综合 Rhoda AI 的研究出版物、CEO 披露及物流和装配场景的独立技术拆解。

[CE001, CE005, CE009, CE010, CE011]

5.2 工业流程、自主牵引力与基准表现

Rhoda AI 的商业部署聚焦经典自动化过去难以完成的高变异仓储和制造任务。在与运动用品零售商 Decathlon 开展的生产评估中,系统端到端自主完成三类流程:轴承拆包、容器拆解和服装退货处理。拆包流程需要双手操作,打开 10 kg 零件箱、取出轴承并分离脆弱塑料内衬;经过 11 小时机器人训练数据后,可自主运行 1.5 小时。容器拆解中,平台经过 17 小时训练数据后,无需人工干预,连续 160 分钟自主清空沉重的 50 磅 Contico 运输箱,并动态适应结构碎片和不断变化的锁扣。服装退货流程中,模型端到端解决视觉歧义和可变形布料配置,借助上下文内模仿,仅通过一次人工视频演示就掌握新服装,无需人工编写子任务脚本。[CE006, CE007, CE008]

工业工作流与部署基准
工业工作流目标用户任务既有做法Rhoda AI 方案已测性能运营限制
轴承倒料汽车零部件搬运手工分拣 / 固定式机器人使用 DVA 双手策略处理可变形袋基于 11h 机器人训练数据自主运行 1.5 hours缺少峰值生产线速度下公开的周期时间基准
Contico 容器拆解仓库大宗收货重型人工箱体搬运(50 lbs)结合空间推理与力适应基于 17h 机器人训练数据连续运行 160 min需要在身体附近搬运 50-lb 物体;能力边界的应变未知
服装退货处理电商服装物流手工拆袋和折叠端到端长上下文视觉记忆通过单次人工示范处理可变形织物每件服装的计算推理成本高
组件处理电子 / 离散制造硬编码轨迹自动化以毫秒级更新的闭环视觉策略完整周期低于 2 分钟,达到工厂标准对极端光照变化和光学遮挡敏感

基于 Decathlon 试点评估、客户概念验证基准和技术报告中的运营数据。

[CE006, CE007, CE008]

5.3 Direct Video Action 架构与运行模式

Rhoda AI 的核心技术突破是 DVA 架构,它把对实体世界的全局理解与低层机器人关节运动学解耦。底层是从零开始用超过一百万条互联网视频预训练的原生因果视频模型,能够捕捉直觉物理、重力和物体互动,无需数百万小时远程操作演示。Rhoda AI 用自研 Context Amortization 目标训练模型,在每个序列位置同时预测未来帧,以数百帧的长上下文视觉记忆,并通过键值缓存保持推理效率。下游执行由紧凑型逆动力学模型负责,它用约 10 小时机身专属遥测,把连续预测视频帧转换为电机扭矩和关节位置。为弥合重型视频模型推理(约 100 毫秒)与实时电机控制回路(10 至 30 毫秒)的延迟差,Rhoda 执行 leapfrog inference,让未来视觉展开与先前动作执行重叠,并依据当前电机状态条件化连续展开,从而保持运动学轨迹连续。[CE002, CE003, CE004, CE012, CE015]

技术与运营架构评估
架构层运营角色技术实现主要依赖技术与运营风险
因果视频预训练世界物理表征基于 1M+ 互联网视频的因果自回归扩散用于大规模网络视频摄取和 FVD 评分的计算集群网络视频视觉分布相对工厂地面环境发生偏移
上下文摊销训练效率扩展使用 KV 缓存进行多帧同步损失预测足够的 GPU VRAM 以支持数百帧序列长时序展开期间的内存压力
逆动力学转换将动作落到硬件上非因果视觉到扭矩回归模型约 10 小时的实体专属运动学遥测数据解耦架构阻止端到端梯度修正
Leapfrog 推理控制器缓解延迟并保持连续性结合动作状态条件的前瞻轨迹规划边缘 GPU 计算上的推理频率低于 100ms动作条件漂移时可能出现随机轨迹跳变

源自 Rhoda AI 架构论文、硬件披露和工程规格。

[CE002, CE003, CE004, CE012]
FE001: 全栈 Direct Video Action 机器人架构

Rhoda AI Direct Video-Action 机器人平台从实体硬件到工业应用的六层架构栈。

工业应用层
包括轴承拆箱倒料、容器拆分和服装退货处理在内的工厂与物流工作流
模块
  • 仓库 AMR/AGV 车队协调
  • 任务执行监控器
Leapfrog 推理控制器
将视频推理延迟(约 100ms)与执行重叠,每 10-30ms 持续发出电机指令
模块
  • 前瞻轨迹规划器
  • 动作连续性调节器
逆动力学模型
非因果动作转换器,将连续预测帧映射为电机扭矩和关节位置
模块
  • 帧到动作回归器
  • 运动学校准模块
因果视频策略引擎
世界模型视频预测器,根据当前相机输入和本体感觉生成未来场景帧
模块
  • 自回归视频 Transformer
  • KV-Cache 内存缓冲区
网络视频预训练流水线
网络规模数据筛选引擎,处理 1M+ 个视频,用于学习物理、运动和 3D 物体恒存性
模块
  • 自动化 CLIP/VLM 筛选器
  • 上下文摊销损失引擎
实体硬件平台
配备轮式底座、双 7-DoF 机械臂和三指灵巧手的移动操作机器人,可承载 25-45 kg 负载
模块
  • 安全认证视觉传感器
  • 故障安全轮刹

分层边界和延迟目标反映已发布的 DVA 架构规格及 CEO 披露。

[CE001, CE002, CE003, CE004, CE015]

5.4 技术差异化、安全控制与尽调风险

Rhoda AI 的架构带来独特的运营和安全优势,也引出相应技术尽调问题。相较黑盒 VLA 策略,DVA 的主要运营优势是视觉可解释性:策略在执行前先以视频展开预测未来状态,工程团队可以检查预测轨迹、确认障碍物净空,并在施加机械力前捕捉故障模式。从系统角度看,轮式移动操作机器人配备故障安全电磁制动器和安全等级视觉传感器,满足工厂地面人机协作标准。但关键风险在于因果视频模型与逆动力学模块之间不可微的边界,这使执行错误无法直接更新世界模型。此外,招聘信号显示 Rhoda 仍高度聚焦网络规模视频筛选、自动过滤流程和 Fréchet Video Distance 评估指标。若要在未测绘的工厂环境中实现无需现场微调的零样本实体执行,公司必须先解决把网络视频先验转化为真实动作的瓶颈;这也是大规模企业部署的主要门槛。[CE010, CE012, CE013, CE014, CE015]

5.5 展品

Chapter 06

06客户

6.1 目标客户细分与买方画像

Rhoda AI 的实体 AI 技术明确面向长期缺工的工业制造和物流企业,而不是消费或服务机器人市场。在目标工厂,商业买方通常是工厂副总裁、自动化负责人、运营工程师和资本配置委员会,按严格的回本周期和吞吐指标评估自动化。目标环境包括汽车零部件子装配线、3PL 配送中心和高产量电商履约中心,这些地方的重复物料处理任务往往人员流失率很高。Rhoda AI 不打算只销售重型设备,而是采用双重商业架构:销售自研移动操作机器人,同时向第三方设备制造商授权 FutureVision 基础模型。这样,运营负责人可以在熟悉的 AGV 形态上部署自主智能,无需改造现有设施;投资人也可按 Labor-as-a-Service 框架测算可观的经常性合同价值。[CU001, CU002, CU007, CU009]

客户分群表
分群买方 / 用户 / 付款方主要用例目标规模收入 / 战略价值尽调缺口
汽车与重型制造工厂运营副总裁 / 自动化工程师 / 资本开支总监轴承倒料、原材料搬运、重型箱体操作每个装配工厂 10–50 台战略价值高;关键任务正常运行时间合同尚未公布汽车客户的公开合同
仓储与合同物流(3PL)设施总经理 / 运营总监 / 损益负责人托盘拆解、分拣、料箱倒料、暂存每个配送中心 50–200 台高规模 LaaS 经常性收入($100k/台/年)缺少已披露的多年期 3PL 商业协议
电商退货与履约履约运营副总裁 / 仓储技术总监 / 损益负责人服装退货拆包、检查、折叠、装箱每个履约中心 20–100 台高周转工作流中的节省人工 ROI 高对新型服装变体的性能退化尚未核实
工业设备 OEM产品副总裁 / 首席技术官 / 企业发展负责人向第三方机器人授权 FutureVision 基础模型跨多个 OEM 的全球机群授权高毛利纯软件授权收入分成尚未宣布 OEM 硬件集成合作

商业分群基于 Rhoda AI 披露和工业机器人市场分析;客户数量及合同金额仍属于未披露的私有数据。

[CU001, CU002, CU003, CU009]

6.2 商业采用轨迹与部署基准

Rhoda AI 的客户采用轨迹一面显示出有希望的技术概念验证基准,一面仍处于商业化早期。公司在一次制造客户初步评估中展示了自主部件处理流程,周期不到两分钟,同时满足严格的生产性能阈值。DVA 架构只需 10 至 20 小时远程操作演示,就能学习复杂的机器人操作任务,绕开传统工业机器人编程所需的数百小时,显著降低采用摩擦。持续运行时间也显示出韧性:容器拆解连续 160 分钟,轴承拆包自主运行 1.5 小时,期间应对照明变化和物品摆放不规则。投资人 Mayfield 模型认为,部署 1,000 台可产生约 $100M ARR,10,000 台时达到 $1B ARR。但这些数字目前只是投资人的前瞻模型,并非经验证的商业车队收入。[CU004, CU005, CU006, CU009]

客户增长 / 采用轨迹表
指标数值日期来源置信度含义缺失分母
组件处理周期时间<2 min / cycle2026-03-10YesPress / AI Insider中超过工业组件生产线的生产 KPI班次期间的总批量和缺陷率
Contico 箱体拆解连续运行160 min autonomous2026-03-10Rhoda AI 技术博客高验证无需停顿的长时间闭环运行跨多日运行班次的长期 MTBF
倒料连续运行时长1.5 hours uncut2026-03-10Rhoda AI 技术博客高展示真实试点设置中的边界情况恢复能力日吞吐量与人工操作员基准的对比
任务适应所需数据10–20 小时远程操作数据2026-03-10Rhoda AI / Mayfield高相比数月工程开发,可实现快速客户上线新 SKU 家族的远程操作人工成本
Mayfield 预计规模层级 11,000 台 / $100M ARR2026-03-10Mayfield 投资论点中隐含每台部署机器人约 $100k 的年度经常性合同当前商业订单储备和已承诺客户数

部署指标反映早期客户概念验证评估和投资人目标模型,而非经审计的生产机群数据。

[CU004, CU005, CU006, CU009]

6.3 客户概念验证与运营流程

Rhoda AI 可公开验证的客户证据集中在技术概念验证(PoC)和真实制造试点,而不是已宣布的生产合同。在一次汽车制造试验中,系统自动完成复杂的双手拆包:从 10 kg 箱中取出轴承、拉出薄包装条并分拣包装材料;行业伙伴此前认为这类任务无法自动化。第二个工业概念验证(PoC)展示了重型 50 磅 Contico 容器的自主拆解,平台在连续 160 分钟试验中准确判断空间、清理碎片并操纵难开的锁扣,无需人工介入。第三次评估探索电商退货处理,平台借助长上下文视觉记忆,端到端自主拆包、检查、折叠和分箱服装。尽管这些试点在不同制造和履约环境中展示了复杂操作能力,Rhoda AI 仍未公布具体客户品牌、商业合同金额或企业部署协议。[CU003, CU004, CU005, CU012, CU013]

已命名客户验证表
客户 / 合作伙伴分群部署 / 用例状态报告结果商业限制
保密一级汽车供应商汽车制造双手轴承倒料、拉片、搬运 10kg 箱体试点概念验证(PoC)自主运行数小时,能够处理边界情况合作伙伴在 Rhoda 试验前认为任务不可行;生产合同尚未公布
保密工业容器搬运商制造业物流拆解 50-lb Contico 容器并清理碎屑试点概念验证(PoC)连续自主运行 160 分钟,无需人工干预单工位试验;全设施多机器人机群经济性尚未验证
保密物流与零售合作伙伴电商履约服装拆包、检查、折叠和装箱评估试点依靠视觉记忆端到端完成整条退货流程商业推出时间表和单位经济数据未披露
保密制造客户零部件制造零部件处理流程,周期不到 2 分钟客户评估在实际工厂环境中超过客户生产 KPI指定参考客户及正式商业条款受 NDA 约束,仍未公开

客户名称对应公开概念验证项目和合作伙伴任务类别;企业客户身份依据商业保密协议保持匿名。

[CU003, CU004, CU005, CU012, CU013]

6.4 客户旅程、留存持久性与采用风险

Rhoda AI 的客户旅程从探索性流程咨询和离线远程操作校准开始,经过现场试点搭建、生产班次验证、多工位车队扩张,最终走向长期软件授权。尽管路径清晰,公司仍面临重大商业风险。CEO Jagdeep Singh 直言,整个机器人行业目前没有任何通用自主机器人在商业现场持续完成有用工作,公开视频演示与可部署企业价值之间存在巨大鸿沟。独立市场分析师还强调,Rhoda AI 必须迅速证明切实投资回报,才能让 $1.7B 投后估值经得起 Tesla Optimus、Boston Dynamics 等资本雄厚竞争者的比较。关键是,截至 2026 年 10 月,Rhoda AI 尚未披露任何具名付费企业参考客户、经验证续约率或 cohort 净留存数据,客户集中度和合同持久性仍是关键尽调缺口。[CU002, CU007, CU008, CU010, CU011, CU013, CU014]

FU001: 客户旅程图

客户采用旅程涵盖探索性视频摄取和远程操作微调,直至试点单元部署、闭环倒料、车队扩张和 OEM 软件授权。

各阶段反映 Rhoda AI 管理层和风险投资支持者所概述的商业概念验证进程及企业部署路线图。

[CU002, CU006, CU007, CU009, CU011]

6.5 展品

Chapter 07

07风险

7.1 监管、安全与数据治理暴露

核心法律与监管问题在于证据不足:指定公开来源无法证明 Rhoda 的认证状态、事故记录、产品责任分配、保险覆盖或网络视频训练的权利框架。这既不代表发生过违约或事故,也不证明从未发生。Rhoda 表示生成式预测可能不一致,产生突兀运动或振荡;独立行业文章则把确定性、容错、安全验证和责任列为采用障碍。公司把视频展开和轮式底盘作为安全辅助,但它们只是设计层面的缓解措施,并非已披露的第三方认证。因此,投资尽调应把生产安全批准和训练数据来源视为尚待满足的先决条件,在判断大规模部署前,由法律顾问和独立机器人安全评估机构审阅底层记录。[CR001, CR003, CR004, CR005, CR012]

监管 / 法律风险登记表
风险领域司法辖区 / 审理地公开状态可能性严重程度已披露的缓解措施剩余风险暴露尽调路径
工业机器人安全与产品责任客户部署所在司法辖区指定来源未能证明已有认证、保险或责任分配中关键视频 rollout 检查、动作调节和轮式平台被描述为设计层面的缓解措施在完成独立测试并通过合同分配责任前,风险较高获取安全论证、危险分析、认证记录、保险凭证、事故日志和客户责任条款
网络视频权利与数据治理来源及部署所在司法辖区已披露网络规模内容筛选,但未披露来源权利和隐私控制中高Rhoda 正在招聘负责摄取、去重、过滤、分类和评估流水线的人员未知,因为内容筛选质量本身不能证明已获许可或符合规定审计数据集谱系、许可、退出和删除流程、隐私审查及赔偿安排
安全验证与保险接受度工业客户现场独立报道将确定性和容错性列为障碍;未指明保险公司或评估机构中高闭环观测及可视化的未来 rollout 可能有助于测试和调试如果保险公司或客户安全团队拒绝随机控制,风险较高委托独立的功能安全审查,并取得保险公司和客户现场的书面接受标准

这是指定证据支持的最高优先级法律与监管尽调领域样本;可能性和严重程度是分析师的定性评估,缺少公开记录属于信息缺口,并不证明不合规。

[CR001, CR003, CR004, CR005, CR012]

7.2 运营可靠性与技术规模

Rhoda 的闭环架构直接应对延迟和轨迹连续性问题,但这套缓解方案也暴露了失效模式。系统必须预测视频、将其转化为动作、执行,再每秒重复数次;Rhoda 采用跳跃式推理、动作条件化和缓存上下文,让这条循环持续运转。公开证据没有披露尾延迟、急停频率、恢复率、功耗或现场平均故障间隔。更根本的是,Rhoda 承认,稳健规划与推理、通过交互自我改进、长时域移动操作,以及高自由度灵巧操作仍是尚未解决的技术挑战。这些缺口使生产可靠性和边缘计算经济性成为可监测风险,而不是已经证实的缺陷:真正决定性的证据,是客户现场在长时间、非筛选运行中的可靠性分布。[CR001, CR002, CR009, CR010, CR012]

运营 / 质量 / 安全风险登记表
失效模式可能性严重程度缓解措施成熟度剩余风险暴露未解决的证据问题
连续视频预测不一致,导致动作不连续或振荡中关键架构层面缓解:动作调节与 leapfrog 推理极端少见条件下的尾部行为尚无公开量化客户现场的急停、碰撞、人工介入和恢复分布
推理延迟或计算中断打断实时控制回路中高架构层面缓解:重叠执行、上下文摊销和 KV 缓存功耗、硬件成本及最坏情况延迟仍未披露机器人端计算物料清单、功耗、p95/p99 延迟及降级模式行为
规划、推理或灵巧性在长时程和移动任务中失效高高早期:公司演示和概念验证运行,并明确承认存在挑战尚未证明能泛化到已披露任务之外针对未见任务、现场、物体及恢复场景的预注册评估
网络视频分布或评估设计给出误导性模型进展信号中高搭建阶段:正在招聘专门负责内容筛选、标注、过滤和基准测试的人员没有公开的基准方法或外部评估数据集构成、污染控制、基准定义及独立复现

各行按潜在严重程度排序。可能性是基于已披露架构和信息缺口的分析评级,并非观测到的事故频率。

[CR001, CR002, CR009, CR010, CR012]

7.3 硬件、合作伙伴与客户依赖

Rhoda 描述了一条异常宽的执行边界,横跨机器人硬件、基础设施、制造扩产、数据整理、评估和基础模型。垂直整合可能提升系统级控制,但也会把执行风险集中在一家年轻公司内部,并带来未披露的供应商、制造、保修和服务依赖。公开验证仍处于早期:Rhoda 报告两项客户概念验证(PoC)任务连续运行数小时且无需干预;其 CEO 称专有硬件已在客户处完成测试,尽管硬件尚未正式发布;投资方 Mayfield 则称一项现场工作流超过了客户 KPI。这些来源既未指明客户,也未证明会形成复购。具名客户参考、部署数量、合同条款,以及从概念验证(PoC)转为付费机队的转化率,才是合适的商业淘汰测试。[CR006, CR007, CR008, CR011, CR016]

合作伙伴 / 依赖风险登记表
依赖项交易对手方角色集中度失效情景严重程度当前缓解措施剩余风险暴露
客户验证与部署转化未具名的工业概念验证合作伙伴验证安全性、可靠性、ROI 和可重复性未知;公开信息没有客户名单或部署分母筛选后的验证项目没有转化为持续付费的车队部署关键两项公司报告的概念验证任务和一项投资者报告的实际工作流在核实具名参考客户、合同和重复订单前,风险较高
集成硬件与制造扩产未披露的零部件、制造和服务供应商构建并支持实体机器人系统公司整体栈层面较高良率、供应商、保修或现场服务限制吞噬资本并延迟 rollout高Rhoda 表示其拥有完整技术栈,并正在投资制造扩产高,因为未披露供应商冗余、成本和服务覆盖
网络视频数据与评估流水线互联网内容来源与内部数据团队提供预训练分布和模型进展衡量方法层面依赖网络规模视频筛选不佳、权利限制或薄弱基准会削弱向机器人控制的迁移高专门开展摄取、去重、过滤、标注和评估工作中高,待完成谱系审查和外部基准复现
融资与估值支持Series A 投资者财团为研发、招聘、工业部署和制造扩产提供资金一轮异常庞大的已披露融资在生产可靠性和单位贡献经济性得到证明前,资本可能已被消耗高$450 million Series A 融资,据报投后估值为 $1.7 billion未知,因为 burn、runway、定价、利润率和营运资金需求均未披露

依赖和集中度评估只使用指定资料池中的公开披露;没有供应商、客户、收入或现金可用性分母。

[CR006, CR007, CR008, CR011, CR012, CR013, CR014, CR016]

7.4 财务模型风险与论点失效标准

$450 million 的 Series A 和据报道 $1.7 billion 的投后估值提供了充足融资缓冲,但已募资金不等于可用现金、现金 runway,也不等于有吸引力的单位经济性证据。来源没有披露现金消耗、毛利率、定价、硬件物料清单、营运资金需求、经常性收入或已签约订单储备。Mayfield 称部署 1,000 台设备可产生约 $100 million 的年度经常性收入,这只是投资方情景假设,不是已报告的 ARR,也不是公司业绩。如果生产部署前仍无法获得独立安全审查;如果长时间客户试验无法复现已发布的演示;如果具名试点无法转化为重复付费设备;或者经审计的贡献经济性无法证明能够覆盖硬件、算力、现场服务和保修成本的可信路径,那么投资论点就应失效。[CR004, CR008, CR011, CR013, CR014, CR015, CR016]

FR001: 剩余风险热力图

仅考虑指定证据中披露的控制措施和证明后的影响、可能性及缓解成熟度定性评估。

截至 2026-10-14,所有单元格均为分析师的定性判断,不是经测量的事故概率、经审计的控制措施或经独立验证的性能评级。

[CR001, CR004, CR009, CR011, CR012, CR013, CR015, CR016]

7.5 展示材料

Chapter 08

08估值

8.1 投资论点与反论点

Rhoda AI 的核心投资论点很有吸引力:它试图解决机器人自动化规模化的首要瓶颈——依赖人工遥操作。公司把专有的 Direct Video Action(DVA)架构引入 FutureVision 平台,利用数亿条公开互联网视频预训练基础模型,先建立对物理规律、物体恒存性和动态交互的稳健世界模型,再用仅 10 至 20 小时的遥操作数据,将这些表征映射到机器人本体。早期在高吞吐汽车装配环境中的评估显示,Rhoda 完成了一项无需人工干预的自主零部件处理工作流,每个循环用时不到两分钟,超过了既有客户基准。公司还做出了明确的系统级硬件取舍:开发配备七自由度机械臂和三指手的轮式移动操作机器人,而不是不稳定的双足人形机器人,优先保证作业安全、动态平衡,并立即兼容现有自动导引车基础设施。 反过来,反论点集中在早期试点演示与生产规模工业部署之间的巨大鸿沟。正如 CEO Jagdeep Singh 坦率承认的那样,尽管具身 AI 领域已投入数十亿美元,如今仍没有通用型自主机器人在商业现场作业中完成有用工作。视频预训练能提供视觉先验,却缺乏触觉本体感知,使系统容易受到多变棕地工厂环境中细微物理分布变化的影响。此外,独立行业分析强调,面对 Boston Dynamics 和 Tesla Optimus 等资金雄厚的竞争者,Rhoda AI $1.7 billion 的投后估值是否对应已证明的投资回报,值得严肃审视。$450 million 的 Series A 资金为算力提供了较长 runway,但随着公司激进扩充研究团队并扩大硬件规模,资本密集度仍然很高。[CV004, CV005, CV006, CV007, CV008, CV012, CV014]

投资论点 / 反论点表
战略支柱投资论点反论点 / 反命题什么会改变判断
模型范式与数据Direct Video Action 利用网络规模视频解决物理常识问题,避开远程操作瓶颈。网络视频缺少本体感觉和触觉反馈;真实工厂中的边缘情况仍会造成高昂失效。第三方基准证明其能在多种制造工作单元中零样本适配。
硬件架构轮式移动操作机器人避开人形机器人的稳定性风险,并能安全接入存量 AMR 设施。轮式形态无法上下楼梯或穿行多层货架,相比人形机器人限制了可覆盖的场地范围。客户审计证明多层场地在无人干预下的周期完成率超过 99%。
商业部署汽车行业评估显示周期不到 2 分钟,为 Rhoda 快速转化客户试点创造条件。单现场评估不能保证多班次可靠性、维护开机率或客户 ROI。具名 Tier-1 制造商的公开案例研究和多年生产服务合同。
市场估值与资本$450M 资金储备提供充足算力 runway,可持续迭代并建立数据飞轮。如果企业采用落后或算力消耗上升,$1.7B 入场估值几乎没有安全边际。披露 ARR 超过 $50M,或车队里程碑超过 500 台可计费生产设备。

对比基础 Physical AI 优势与非结构化工业环境中运营扩张阻力的战略支柱。

[CV004, CV005, CV006, CV007, CV008, CV012]

8.2 融资背景、估值倍数与行业可比公司

2026 年 3 月,Rhoda AI 结束了 18 个月的隐身运营,宣布完成 $450 million 的 Series A,投后估值为 $1.7 billion。本轮融资由 Premji Invest 领投,Temasek、Khosla Ventures、Mayfield、Matter Venture Partners、Capricorn Investment Group、Prelude Ventures 和 John Doerr 等一线投资方组成的财团参与。这是机器人历史上规模最大的 Series A 之一,反映出资本高度集中涌向物理基础模型。该轮融资恰逢 2026 年 1 月和 2 月前所未有的募资浪潮:期间 AI 机器人领域的风险投资在超过 110 笔交易中突破 $4.3 billion,预计全年行业流入将达到 $20 billion。 从估值可比角度看,Rhoda AI 的 $1.7 billion 估值使其跻身顶级私人物理 AI 独角兽之列,不过商业化阶段更早。相比之下,Skild AI 在 2026 年 1 月完成 $1.4 billion 的 Series C 后估值达到 $14 billion;Apptronik 通过增资将其人形机器人平台估值推高至 $5 billion 以上;Figure AI 的估值则达到 $39 billion。然而,与已公开企业部署情况的后期同行不同,Rhoda AI 尚未披露经常性软件收入、年度经常性收入 run-rate 或可计费机队客户数量。由于 Rhoda AI 的 $1.7 billion 估值完全建立在未经验证的商业收入之上的远期倍数扩张,行业观察人士指出,公司必须迅速证明商业客户的可验证投资回报,以免未来融资轮次出现倍数压缩。[CV001, CV002, CV003, CV008, CV009, CV010, CV013]

建议总结表
指标 / 维度评估证据 / 驱动因素决策含义
总体建议research-more突破性视频基础模型的优势,被尚未验证的车队规模单位经济性抵消在多客户部署得到验证前,暂缓配置资本。
信心水平medium已核实 $450M 融资及一线财团,但多现场客户遥测尚未核实谨慎评估,避免激进建仓。
风险评级high硬件 / 模型扩张资本密集、面临超大市值 AI 实验室竞争,且存在现场失效模式要求高优先级清算优先权保护,或在二级转让中获得折扣。
估值立场stretchedSeries A 的 $1.7B 投后估值,实质上是对商业化前收入给出的纯企业价值倍数在商业风险显著降低前,普通股的风险收益比缺乏吸引力。
资本与悬垂$450M Series APremji Invest、Temasek 和 Khosla 财团形成庞大的新发优先股堆叠在下行情景出售中,必须先建立较厚的下行缓冲,股权价值才会落到普通股股东手中。

估值立场和风险评级综合考虑 Series A 定价、Physical AI 行业倍数,以及缺少公开商业车队审计数据。

[CV001, CV002, CV008, CV010, CV013]

8.3 情景分析:乐观、基准与悲观回报画像

为评估 Rhoda AI 的风险调整后回报画像,我们在三至五年的投资期限内建立三种不同的商业采用情景。在乐观情景(概率 25%)下,Rhoda AI 成功把当前的制造和物流评估转化为多工厂商业部署。财团合作方 Mayfield 推广的投资判断假设认为,在 Labor-as-a-Service 模式下部署 1,000 台移动操作机器人,可从 $30 trillion 的全球体力劳动市场中带来约 $100 million 的年度经常性收入,并最终在 10,000 台设备时达到十亿美元级 run-rate。在这一情景中,Rhoda AI 成为工业操作领域的市场领导者,估值扩大到 $5 billion 至 $8 billion,为 Series A 投资者带来 3x 至 5x 回报。 在基准情景(概率 50%)下,Rhoda AI 稳步但审慎地推进采用,面对标准工业采购周期、工厂集成障碍和广泛的安全认证。到 2028 年,公司在物流退货处理和结构化制造单元中部署 150 至 300 台设备,硬件与模型授权合计产生 $15 million 至 $30 million 收入。增长放缓和行业整体倍数正常化会将远期企业价值倍数压缩到符合市场的区间,估值预计为 $1.5 billion 至 $2.2 billion,资本回报持平或小幅增长。在悲观情景(概率 25%)下,模型泛化在真实生产线的意外边缘案例面前失灵,导致循环中断和客户流失。持续训练视频基础模型带来的高额算力消耗会耗尽 $450 million 资金,却仍未达到单位经济性盈亏平衡,最终触发一轮低价融资或 $400 million 至 $700 million 的结构化资本重组;其中高级清算优先权会抹去普通股权益。[CV001, CV007, CV008, CV010, CV011, CV012]

乐观 / 基准 / 悲观情景表
情景核心假设估值 / 回报逻辑主要下行风险概率信号
乐观情景到 2028 年,汽车 / 物流试点快速转化为 1,000+ 台已部署移动操作机器人;Labor-as-a-Service 模式下经常性收入达到 $100M+。估值扩张至 $5B-$8B(远期 ARR 倍数 15x-20x),为 Series A 投资者带来 3x-5x 回报。算力通胀、硬件保修成本或客户采购周期延迟。25%(需要商业执行和硬件可靠性都无瑕疵)
基准情景温和采用,在结构化物流工作流中部署 150-300 台;混合模式授权和自有设备带来 $15M-$30M 收入。估值稳定在 $1.5B-$2.2B(增长放缓压缩远期倍数);回报为 1.0x-1.3x,后续融资可能持平。客户支持开销和定制集成要求压缩利润率。50%(考虑到存量工厂集成摩擦,这是最可能的轨迹)
悲观情景非结构化环境出现技术分布偏移;试点流失;高算力消耗耗尽 $450M 资金,却未达到可规模化的单位经济性。在下轮融资或困境资产出售中,估值压缩至 $400M-$700M;优先股拿走全部回收价值,普通股归零。硬件安全事故、模型幻觉导致工厂停机,或出现更优秀的开源机器人模型。25%(早期深科技机器人面临较高资本密集型风险)

基于车队部署规模、软件抽成率和终值倍数假设的情景回报建模。

[CV001, CV007, CV008, CV010, CV011, CV012]

8.4 建议逻辑、淘汰标准与最终尽调要求

综合技术、竞争和估值证据后,整体尽调建议为继续研究,风险评级为高,估值判断偏高。Rhoda AI 的闭环视频预测控制架构、连续创业深科技创业者 Jagdeep Singh 领导的资深团队,以及资本充足的投资财团,构成了异常扎实的基础实力。但在尚未独立验证多现场 uptime、部署规模和客户合同续签前,就为一家早期 Series A 公司支付 $1.7 billion,达不到防御性资本配置所需的标准价格纪律。 在把 Rhoda AI 上调为可执行的买入判断前,机构投资者应监测明确的论点失效触发条件,并执行具体的尽调淘汰标准。关键淘汰标准包括:(1)当前汽车制造和物流试点未能在 12 个月内转化为多年期商业合同;(2)真实工厂班次中模型性能过度下降,或与安全相关的急停频率过高;(3)算力消耗率上升,使运营现金 runway 降至 24 个月以下。确认性尽调必须要求机密访问客户参考访谈、经审计的单位级贡献利润,以及 Series A 资本结构表,以评估清算优先权的悬置影响和高级结构性权利。[CV001, CV005, CV006, CV008, CV010, CV013, CV014]

FV001: 推荐逻辑

从物理基础模型的验证结果,到商业化扩张瓶颈,再到由此得出的继续研究建议,这是一条完整的逻辑链。

节点 / 连接节点 / 起点终点说明
节点 1 基础 DVA 模型与 $450M A 轮资金池 [N1]
节点 2 高吞吐汽车试点成功(周期不到 2 分钟) [N2]
节点 3 务实的轮式移动操作机器人硬件设计 [N3]
节点 4 商业瓶颈:多站点 TCO 与车队规模尚未验证 [N4]
节点 5 估值审视:$1.7B 价格要求企业立即产生 ROI [N5]
节点 6 尽调结论:继续研究 / 估值偏高 [N6]
连接 1 基础 DVA 模型与 $450M A 轮资金池 [N1]高吞吐汽车试点成功(周期不到 2 分钟) [N2]支持试点演示
连接 2 高吞吐汽车试点成功(周期不到 2 分钟) [N2]务实的轮式移动操作机器人硬件设计 [N3]为硬件形态提供依据
连接 3 务实的轮式移动操作机器人硬件设计 [N3]商业瓶颈:多站点 TCO 与车队规模尚未验证 [N4]面临工厂现场阻力
连接 4 商业瓶颈:多站点 TCO 与车队规模尚未验证 [N4]估值审视:$1.7B 价格要求企业立即产生 ROI [N5]放大估值风险
连接 5 估值审视:$1.7B 价格要求企业立即产生 ROI [N5]尽调结论:继续研究 / 估值偏高 [N6]推导出谨慎结论
[CV001, CV005, CV006, CV008, CV010, CV013, CV014]

8.5 展示材料

免责声明

本报告是基于公开证据的尽调快照,不构成投资建议。重要的财务、法律、技术和合同事实仍未公开,任何投资决策前都应直接向管理层及原始文件核实。

证据索引

结论
编号陈述可信度来源
CO001 Rhoda AI raised $450 million in Series A equity funding in March 2026 at a post-money valuation of $1.7 billion. 高 SO004, SO005, SO007
CO002 Rhoda AI's Series A round was led by Premji Invest with participation from Khosla Ventures, Temasek, Mayfield, Capricorn Investment Group, and John Doerr. 高 SO004, SO007, SO008
CO003 Rhoda AI was founded in late 2024 and operated in stealth development for approximately 18 months before its public launch. 中 SO007, SO009
CO004 Rhoda AI's leadership includes CEO Jagdeep Singh, Chief Science Officer Eric Ryan Chan, and Stanford Professor Gordon Wetzstein. 高 SO004, SO009
CO005 Rhoda AI is headquartered in Palo Alto, California, developing foundation models for real-world industrial robotic manipulation. 高 SO001, SO002, SO007
CO006 Rhoda AI targets industrial and commercial manipulation tasks across automotive, manufacturing, logistics, and ecommerce sectors. 高 SO001, SO007
CO007 Rhoda AI's FutureVision platform uses a Direct Video Action architecture that formulates robot control as a continuous video-predictive task. 高 SO001, SO005, SO009
CO008 FutureVision pretrains models on hundreds of millions of internet videos before fine-tuning with 1 to 10 hours of robot trajectory data. 高 SO001, SO007, SO009
CO009 Rhoda AI demonstrated an autonomous component-processing manufacturing workflow completed in under two minutes per cycle without human intervention. 中 SO003, SO009, SO010
CO010 Venture backer Mayfield characterized Rhoda AI's business opportunity as Labor as a Service addressing a $30 trillion manual labor market. 中 SO007, SO010
CO011 Industry analysts caution that video-predictive models face steep computational costs, edge-hardware latency limits, and unproven safety frameworks. 中 SO003
CO012 Rhoda AI has not disclosed commercial revenue run-rates, annualized recurring revenue (ARR), or gross margin metrics as of October 2026. 中 SO003, SO004, SO007
CO013 Rhoda AI is developing proprietary wheeled mobile manipulators with 7-DoF arms and dexterous hands rather than bipedal humanoids. 高 SO006, SO007
CO014 Rhoda AI intends to commercialize both proprietary robots and software platform licensing across multiple robotic hardware vendors. 中 SO007, SO008
CO015 Rhoda AI has cited partnerships with leading industrial partners but has not publicly named paying reference customers. 中 SO003
CM001 Rhoda AI raised $450 million in Series A and Series B funding rounds, valuing the physical AI robotics company at $1.7 billion as it exited stealth in March 2026. 高 SM001, SM002, SM008
CM002 Rhoda AI's Direct Video Action architecture pretrains foundation models on hundreds of millions of internet videos to learn motion and intuitive physics before fine-tuning with 10 to 20 hours of robot teleoperation data. 高 SM001, SM002, SM003
CM003 Mayfield Fund frames the target addressable market for physical AI as a $30 trillion annual global manual labor market, targeting repetitive tasks rather than traditional narrow robotics automation. 中 SM004
CM004 Under Mayfield's Labor-as-a-Service model projections, deploying 1,000 robotic units would generate approximately $100 million in annual recurring revenue, and 10,000 units would generate $1 billion in ARR. 中 SM004
CM005 In live manufacturing evaluations with an automotive partner, Rhoda AI demonstrated autonomous completion of a component-processing workflow in under two minutes per cycle without human intervention. 高 SM002, SM003
CM006 Industrial buyers in manufacturing and logistics express operational preference for wheeled mobile manipulators over bipedal humanoids due to lower fall risks, existing AMR facility compatibility, and reliable e-stop safety. 中 SM005
CM007 Rhoda AI has designed its mobile manipulator hardware to handle payloads up to 45 kilograms close to the body, utilizing seven-degree-of-freedom arms and three-fingered hands to balance strength with dexterity. 中 SM005
CM008 The broader AI robotics sector experienced over $4.3 billion raised across 110 funding rounds in January 2026, creating intense capital competition among foundation model providers like Skild AI, Apptronik, and Figure AI. 中 SM006
CM009 Industry analysts caution that premium early valuations like Rhoda AI's $1.7 billion face scrutiny unless startups prove near-term enterprise ROI and solve safety and data-bias hurdles in dynamic production environments. 中 SM006
CM010 CEO Jagdeep Singh observed that despite billions invested across more than 100 humanoid ventures, no generalist autonomous robot has achieved reliable production deployment in the field. 中 SM005
CM011 Rhoda AI pursues a dual commercial strategy comprising proprietary mobile manipulator robot hardware alongside software licensing of its FutureVision foundation model to third-party manufacturers. 高 SM002, SM006, SM007
CM012 Automotive assembly tasks targeted by Rhoda AI include decanting 10-kilogram parts boxes, manipulating straps and deformable bags, and clearing debris from 50-pound Contico containers. 中 SM001, SM006
CM013 Secondary trading and pricing data for Rhoda AI on private exchanges remain limited, with post-money Series A-2 filings establishing a $35.17 per-share benchmark across venture tranches. 中 SM008
CM014 Commercial terms, customer software take rates, gross margin targets, and paid renewal metrics for Rhoda AI's early automotive pilots remain undisclosed in official announcements and secondary market reports. 中 SM002, SM008
CP001 The physical AI and robotics automation landscape is bifurcated between hardware-agnostic foundation model pure-plays, vertically integrated humanoid OEMs, and legacy industrial robot manufacturers. 中 SP006, SP008
CP002 Skild AI secured $1.4B in Series C funding at a $14B valuation in January 2026 to scale robot-agnostic foundational AI models across diverse hardware platforms. 中 SP006
CP003 Figure AI reached a $39B valuation with over $1B raised by early 2026, while Apptronik secured a $520M Series A extension to reach a valuation exceeding $5B. 中 SP006
CP004 Traditional industrial robotics incumbents like FANUC, ABB, and KUKA maintain market dominance in structured factory cells through deterministic trajectory programming but lack adaptive generative perception. 中 SP002, SP003
CP005 Mainstream Vision-Language-Action (VLA) architectures often struggle in dynamic industrial environments due to limited visual context memory and high teleoperation data requirements. 中 SP002, SP008
CP006 Rhoda's Direct Video-Action architecture executes closed-loop video predictive control updated every few hundred milliseconds, contrasting with open-loop planning. 高 SP002, SP004
CP007 Rhoda opted for a wheeled mobile manipulator with seven degree-of-freedom arms and 45kg payload capacity over bipedal legs to avoid dynamic balancing failure modes and emergency stop collapse. 中 SP005
CP008 Rhoda claims FutureVision enables robots to learn new manipulation tasks with roughly ten hours of robot teleoperation data after internet-scale video pretraining. 高 SP001, SP002, SP003
CP009 Rhoda CEO Jagdeep Singh stated that models trained on less curated internet video performed better by learning intuitive physical motion priors than models trained strictly on task demonstrations. 中 SP004
CP010 Rhoda completed an autonomous component-processing manufacturing evaluation in under two minutes per cycle without human intervention. 高 SP002, SP007
CP011 Unlike pure-play foundation models that deploy across multi-client hardware from inception, Rhoda's commercial sequence prioritizes in-house hardware verification before licensing FutureVision to third parties. 中 SP008
CP012 Competitors in the foundation model robotics sector have not published standardized public API pricing or unit costs, relying instead on bespoke enterprise pilots. 中 SP001, SP008
CP013 Rhoda raised a $450M Series A round at a $1.7B post-money valuation led by Premji Invest with participation from Khosla Ventures and Temasek. 中 SP001, SP006
CP014 Enterprise customers face significant switching costs when transitioning from fixed industrial workcells due to plant safety certifications and existing AMR/AGV infrastructure investments. 中 SP001, SP005
CP015 Rhoda plans a dual revenue structure encompassing direct deployment of proprietary mobile manipulators and software licensing of FutureVision to third-party robotics manufacturers. 高 SP002, SP006
CP016 Public launch materials for Rhoda AI lack callable API endpoints, developer documentation, or SDK job controls, restricting near-term access to gated enterprise pilots. 中 SP001
CI001 Rhoda AI secured $450 million in Series A funding to develop its robot foundation models upon exiting stealth in March 2026. 高 SI001, SI003
CI002 Rhoda AI's Series A financing was completed at a post-money valuation of $1.7 billion. 中 SI001, SI002
CI003 The company's commercial monetization model centers on licensing its FutureVision intelligence platform to third-party robotics hardware and software makers. 高 SI002, SI003
CI004 Rhoda is concurrently pursuing a full-stack robotics solution, developing proprietary mobile manipulators with dexterous hands on wheeled bases. 中 SI004, SI005
CI005 The company's proprietary hardware design targets a 25-kilogram continuous payload and up to 40-kilogram peak handling capacity. 中 SI005
CI006 In an early manufacturing pilot evaluation, Rhoda's system autonomously completed a component-processing workflow in under two minutes per cycle. 高 SI002, SI003
CI007 Investor Mayfield models a Labor-as-a-Service economics framework where 1,000 deployed robotic units generate approximately $100 million in recurring annual revenue. 中 SI008
CI008 Under Mayfield's projected deployment model, scaling to 10,000 active robotic units represents a $1 billion annual recurring revenue business. 中 SI008
CI009 Rhoda's Direct Video Action architecture demonstrates data efficiency by acquiring new manipulation tasks with as little as 1 to 10 hours of robot teleoperation data. 中 SI001, SI006
CI010 Rhoda employed over 100 team members upon exiting stealth and is investing aggressively in compute infrastructure, model research, and manufacturing scale-up. 中 SI001, SI007
CI011 Industry analysts highlight that high valuations in AI robotics invite scrutiny, requiring Rhoda to demonstrate rapid commercial return on investment amid competition from Tesla and Boston Dynamics. 中 SI005
CI012 Market analysts note that significant obstacles including operational reliability, safety certification, and deployment costs must be resolved before general-purpose robots achieve broad commercial adoption. 中 SI009
CI013 Rhoda remains pre-revenue with zero disclosed commercial customer contracts, deployments, or verified recurring revenue figures. 中 SI001, SI005
CI014 Hardware capital expenditure, maintenance amortizations, and inference compute represent primary gross margin headwinds for full-stack robotics delivery. 中 SI004, SI008
CI015 CEO Jagdeep Singh observed that the majority of existing humanoid robotics systems function as fragile prototypes unable to handle sustained industrial duty cycles reliably. 中 SI004
CE001 Rhoda AI develops the Direct Video-Action (DVA) architecture to formulate robotic manipulation control as causal video prediction. 高 SE001, SE003
CE002 The DVA causal video model is pre-trained from scratch on over one million internet-scale videos using a context amortization objective. 高 SE003, SE004
CE003 An inverse dynamics model translates consecutive predicted video frames into motor commands and end-effector actions using approximately ten hours of embodiment data. 高 SE003, SE004
CE004 Leapfrog inference overlaps video model computation latency with the execution of previously generated actions to maintain continuous real-time robot control. 高 SE003, SE004
CE005 Rhoda AI's FutureVision platform deploys closed-loop video predictive control to adapt robot actions to dynamic variations in materials and layouts. 中 SE002, SE007
CE006 In automotive and logistics tasks, Rhoda AI demonstrated autonomous decanting for 1.5 hours and container breakdown for 160 continuous minutes without human intervention. 高 SE003, SE004
CE007 Long-context visual memory spanning hundreds of video frames enables end-to-end clothing returns processing without hand-engineered subtask scaffolding. 中 SE003
CE008 One-shot in-context learning enables robots to replicate novel item sorting and stroke-ordered drawing from single human video demonstrations without fine-tuning weights. 高 SE003, SE004
CE009 Rhoda AI developed a proprietary mobile manipulator hardware platform featuring a wheeled base, seven-degree-of-freedom arms, and dexterous three-fingered hands. 中 SE005
CE010 The wheeled base architecture was selected over bipedal humanoid legs to ensure tip-over stability, emergency-stop safety, and compatibility with existing warehouse AGVs. 中 SE005
CE011 Rhoda AI's hardware platform is engineered for a 25-kilogram continuous payload rating, up to 45 kilograms close to the body, with three-year continuous operational durability. 中 SE005, SE006
CE012 Unlike vision-language-action policies that are end-to-end differentiable, the decoupled boundary between DVA's video model and inverse dynamics model prevents downstream motor errors from directly updating the world model. 中 SE004
CE013 Engineering recruiting criteria emphasize web-scale video ingestion, automated CLIP-based curation, and Fréchet Video Distance benchmarks rather than teleoperation fleet expansion. 中 SE008
CE014 Industrial deployment remains gated by real-world factory safety validation, unstructured scene noise, and on-site task fine-tuning bottlenecks. 中 SE001
CE015 Autonomous video rollout generation enables policy interpretability and counterfactual safety verification before motor actions are committed to hardware. 中 SE003
CU001 Rhoda AI targets manufacturing and logistics enterprises as its primary customer vertical for physical AI automation. 中 SU001, SU008
CU002 Rhoda AI plans a dual commercial model encompassing proprietary robotic mobile manipulators and software licensing of its FutureVision foundation model to third-party equipment makers. 中 SU005, SU006
CU003 Rhoda AI deployed proof-of-concept systems with industrial partners to automate complex bimanual decanting of automotive bearings from boxes into totes. 高 SU003, SU005
CU004 Rhoda AI demonstrated an autonomous Contico container breakdown task managing 50-pound boxes and debris for 160 continuous minutes without human intervention. 中 SU003
CU005 In a manufacturing customer evaluation, Rhoda AI's system completed an autonomous component-processing workflow in under two minutes per cycle while meeting production benchmarks. 中 SU002, SU006
CU006 Rhoda AI's Direct Video Action architecture enables robots to acquire complex real-world manipulation tasks with 10 to 20 hours of robot demonstration data. 高 SU003, SU008
CU007 Rhoda AI chose a wheeled mobile base rather than bipedal humanoid legs to ensure balance stability, emergency-stop safety, and rapid customer integration in facilities with existing AGVs. 中 SU004
CU008 CEO Jagdeep Singh stated that despite heavy industry funding and viral video demos, no generalist autonomous robot is currently performing useful work in commercial field operations today. 中 SU004
CU009 Investor Mayfield models that deploying 1,000 robotic units could generate approximately $100 million in recurring annual revenue, scaling to over $1 billion ARR at 10,000 units. 中 SU002, SU008
CU010 Market analysts note that Rhoda AI must quickly demonstrate real-world return on investment to justify its $1.7 billion valuation amidst competition from Tesla Optimus and Boston Dynamics. 中 SU005
CU011 Rhoda AI owns the full robotics stack including high-performance proprietary mobile hardware and foundation world models to support commercial deployment scale-up. 中 SU004, SU007
CU012 Rhoda AI evaluated returns processing involving garment un-packaging, inspection, folding, and binning end-to-end using long-context visual memory without hand-engineered subtask scaffolding. 中 SU003
CU013 As of October 2026, Rhoda AI has disclosed zero named paid enterprise customers, verified multi-unit production contracts, or public reference customer logos. 中 SU001, SU004, SU005
CU014 Historical customer retention metrics, net revenue retention rates, contract duration terms, and cohort churn figures remain completely undisclosed private data. 中 SU001, SU005, SU008
CR001 Rhoda says consecutive stochastic video predictions can become inconsistent and lead to jerky robot movement or oscillation. 中 SR003
CR002 Rhoda says leapfrog inference overlaps prediction with action execution and conditions each prediction on the action currently being executed to maintain trajectory continuity. 中 SR003
CR003 Rhoda presents autoregressive video rollouts as a way to debug, compare, and inspect robot-policy behavior, including for safety evaluation. 中 SR003
CR004 AmericaBots reports that Rhoda has not addressed the liability and safety-validation frameworks required for autonomous manufacturing systems. 中 SR009
CR005 Rhoda's CEO says the wheeled platform avoids a bipedal robot collapsing when emergency power is cut, reducing one specific safety hazard. 中 SR004
CR006 The Association for Advancing Automation reports that Rhoda was testing proprietary hardware with customers before formally announcing that hardware. 中 SR004
CR007 Rhoda says two industrial customer proof-of-concept tasks operated successfully for multiple hours without human intervention. 中 SR003
CR008 Investor Mayfield claims Rhoda completed a live manufacturing component-processing workflow in under two minutes per cycle without human intervention and exceeded customer KPIs. 低 SR008
CR009 Rhoda identifies robust planning and reasoning, self-improvement through interaction, long-horizon mobile manipulation, and high-degree-of-freedom dexterous control as major remaining challenges. 中 SR003
CR010 Independent technical commentary also characterizes planning, interaction-driven improvement, and precise long-horizon mobile manipulation as unresolved challenges for Rhoda. 中 SR001
CR011 Rhoda's reposted job description says the company owns the stack from robot hardware and infrastructure through foundation models and is investing in hardware development and manufacturing scale-up. 中 SR007
CR012 Rhoda's reposted job description says its web-video curation, annotation, and evaluation work directly determines pretraining-distribution quality and how clearly model progress can be measured. 中 SR007
CR013 The AI Insider reports Rhoda's statement that its funding will support research, engineering, industrial deployment expansion, and technical-team growth. 中 SR006
CR014 SuperbCrew reports that Rhoda raised $450 million in Series A equity financing at a $1.7 billion post-money valuation. 中 SR005
CR015 Mayfield projects that 1,000 deployed Rhoda units would generate about $100 million in recurring annual revenue and that 10,000 units would produce $1 billion of ARR. 低 SR008
CR016 AmericaBots reports that Rhoda has not named the industrial partners behind its claimed partnerships. 中 SR009
CV001 Rhoda AI's $1.7 billion entry valuation is not anchored by any disclosed recurring revenue, backlog, or fleet-economics baseline. 高 SV001, SV003
CV002 The Series A round was led by Premji Invest, with participation from Temasek, Khosla Ventures, Mayfield, Matter Venture Partners, Capricorn Investment Group, Prelude Ventures, and John Doerr. 高 SV001, SV003
CV003 Rhoda AI operated in stealth mode for 18 months prior to its public launch and Series A announcement in March 2026. 中 SV001, SV004
CV004 Rhoda AI's Direct Video Action platform pretrains on hundreds of millions of public internet videos to establish motion priors before fine-tuning with 10 to 20 hours of robot teleoperation data. 中 SV001, SV007
CV005 Rhoda AI demonstrated an autonomous component-processing workflow completed in under two minutes per cycle without human intervention during a high-volume manufacturing evaluation. 高 SV003, SV004
CV006 Rhoda AI selected a wheeled mobile manipulator configuration with seven-degree-of-freedom arms and three-fingered hands over bipedal humanoid legs to ensure dynamic stability, payload capacity, and factory safety. 中 SV005
CV007 CEO Jagdeep Singh observed that despite billions of dollars raised and promotional videos published across the industry, no generalist autonomous robot is performing useful work in commercial field operations today. 中 SV005
CV008 Independent market reporting highlighted that Rhoda AI's $1.7 billion valuation invites critical scrutiny regarding real-world return on investment amid competition from Tesla Optimus and Boston Dynamics. 中 SV006
CV009 Sector comparable financings in early 2026 established high private marks, including Skild AI at $14 billion, Apptronik at over $5 billion, and Figure AI at $39 billion. 中 SV006
CV010 Industry analysis notes that independently verified performance, multi-shift reliability, and total cost of ownership across diverse customer sites have yet to emerge publicly for Rhoda AI. 中 SV002
CV011 Mayfield's investment thesis projects that deploying 1,000 units could yield approximately $100 million in annual recurring revenue under a Labor-as-a-Service model targeting the $30 trillion manual labor economy. 中 SV009
CV012 Rhoda AI is investing capital into expanding its multidisciplinary team and research infrastructure across model engineering and hardware scale-up. 高 SV003, SV008
CV013 The diligence recommendation for Rhoda AI is research-more with high risk and a stretched valuation stance, requiring verified customer fleet deployments before underwriting follow-on exposure. 中 SV002, SV006
CV014 Closed-loop video predictive control repeating every few hundred milliseconds enables real-time trajectory adjustments that distinguish Rhoda AI from open-loop planning systems. 中 SV001, SV004
来源
编号出版方标题引文
SO001 Rhoda AI Rhoda AI We first pre-train our model with hundreds of millions of videos, giving it a strong prior on motion, physics, and dynamics. We then post-train on robot data with actions to teach the model to learn specific tasks.
SO002 Rhoda AI Contact | Rhoda AI Rhoda develops cutting edge robotic intelligence to bring generalist robots to the real world.
SO003 AmericaBots Rhoda AI exits stealth with $450M to train robots from video · AmericaBots The skeptic’s case: Stealth-phase demos are typically cherry-picked scenarios optimized for the technology. The real test is whether Rhoda’s system handles edge cases, manufacturing anomalies, and customer-specific workflows at scale.
SO004 Investing.com Rhoda AI raises $450M to train robots using online videos By Investing.com The round was led by Premji Invest and values the startup at $1.7 billion, including the money raised.
SO005 IndexBox Rhoda AI FutureVision Platform & $450M Funding at $1.7B Valuation - News and Statistics - IndexBox Rhoda AI has secured 450 million dollars in a Series A financing round, according to a Reuters report. The investment values the company at 1.7 billion dollars.
SO006 Association for Advancing Automation Rhoda AI's CEO on Building Mass Producing Humanoids “You might be surprised to hear this,” says Rhoda AI CEO, Jagdeep Singh, “but we think that we could be the first autonomous manipulation capable robot that's deployed at scale.”
SO007 SuperbCrew Rhoda AI Raises $450 Million In Series A Funding Round - SuperbCrew Rhoda AI, a Palo Alto-based startup specializing in foundational AI models for industrial robotics, has secured $450 million in a Series A funding round. This investment values the company at $1.7 billion post money, marking a substantial entry into the market as it emerges from 18 months of stealth development.
SO008 The Star Rhoda AI raises $450 million at $1.7 billion valuation, unveils robot intelligence platform The company expects to eventually license FutureVision to companies running robotic hardware and software platforms.
SO009 The AI Insider Rhoda AI Exits Stealth with $450M Series A Funding Round to Scale Robitics Intelligence Platform Along with Singh, a technology entrepreneur who has previously founded and scaled multiple deep-technology companies, the leadership team also includes Chief Science Officer Eric Ryan Chan, a researcher in computer vision and generative modeling who previously worked on generative systems at WorldLabs, and Gordon Wetzstein, a Stanford University professor and head of the Computational Imaging Lab.
SO010 Mayfield Why We Backed Rhoda AI: The $30 Trillion Physical AI Market Rhoda has already demonstrated autonomous operation in live manufacturing environments: a component-processing workflow completed in under two minutes per cycle, with no human intervention, exceeding customer KPIs.
SM001 Humanoids Daily Rhoda AI Hits $1.7B Valuation, Unveils "Direct Video-Action" Model to Bridge the Real-World Gap By contrast, Rhoda claims its model can learn complex, long-horizon industrial tasks with as little as 10 to 20 hours of specific robot data, relying on its pre-trained "world model" to handle the messy variables of reality.
SM002 Rhoda AI Press Release | Rhoda AI In a recent high-volume manufacturing evaluation, Rhoda completed a component-processing workflow in under two minutes per cycle without human intervention, exceeding customer KPIs.
SM003 SiliconANGLE Rhoda AI raises $450M to build foundational robotics models that learn from internet videos The company added that it has run successful tests using off-the-shelf parts with an automotive firm and has plans to license its AI model to customers.
SM004 Mayfield Fund Why We Backed Rhoda AI: The $30 Trillion Physical AI Market The global market for manual labor is roughly $30 trillion annually. Even fractional penetration of that market is a business of staggering scale. At 1,000 deployed units, Rhoda generates ~$100M in recurring annual revenue. At 10,000 units, it’s a billion-dollar ARR business.
SM005 Association for Advancing Automation Rhoda AI's CEO on Building Mass Producing Humanoids The gap between the hype and the reality is is is quite substantial. We just did not want to be yet another company with just demos.
SM006 SuperbCrew Rhoda AI Raises $450 Million In Series A Funding Round - SuperbCrew High valuations in the sector invite scrutiny; Rhoda must demonstrate real world ROI quickly to justify $1.7 billion amid competition from established players like Boston Dynamics and Tesla’s Optimus.
SM007 The AI Insider Rhoda AI Exits Stealth with $450M Series A Funding Round to Scale Robitics Intelligence Platform Rhoda AI said it plans to license the technology to partners using a range of hardware and software platforms.
SM008 Forge Global Rhoda AI IPO Timeline and Financing Details - Forge Forge Price may rely on a very limited number of inputs in its calculation. Forge Price is prepared and disseminated solely for informational purposes.
SP001 COEY Rhoda AI’s Direct Video-Action Model Wants to Make Robots “Web-Trained” and Factory-Ready · COEY As of today, Rhoda is positioning DVA and its broader platform as production-oriented, but public API documentation is not part of the launch materials.
SP002 Rhoda AI Press Release | Rhoda AI Built on this architecture, 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.
SP003 Pulse 2.0 Rhoda AI: $450 Million Series A Raised To Deploy Video-Predictive Robots In Real-World Environments In many cases, Rhoda says its system can learn new behaviors with roughly ten hours of teleoperation data.
SP004 Association for Advancing Automation (A3) Rhoda AI is Rethinking How We Train Robots Rather than planning out the robot’s actions in, say, five second bursts, Rhoda’s closed loop system keeps predictions to a few hundred milliseconds at a time, meaning it can respond more quickly to the inevitable variations in real-world settings.
SP005 Association for Advancing Automation (A3) Rhoda AI's CEO on Building Mass Producing Humanoids If we have to e-stop the robot, we don't have to worry about the robot collapsing and either damaging itself or worse injuring a human being
SP006 SuperbCrew Rhoda AI Raises $450 Million In Series A Funding Round - SuperbCrew Skild AI raised $1.4 billion in Series C at a $14 billion valuation in January 2026, focusing on robot agnostic AI models... Apptronik secured a $520 million Series A extension in February 2026, bringing its total to $935 million at over $5 billion valuation, for humanoid robots. Figure AI achieved a $39 billion valuation with over $1 billion raised by early 2026, emphasizing enterprise humanoid deployments.
SP007 The AI Insider Rhoda AI Exits Stealth with $450M Series A Funding Round to Scale Robitics Intelligence Platform Rhoda AI said its Direct Video Action architecture integrates perception and control in a closed-loop system that allows robots to update actions continuously as conditions change, rather than relying on fixed plans that can fail in dynamic environments.
SP008 Note (生成AI事例集) Rhoda AI: A Foundation Model for Predicting Future Video and Converting It into Robot Actions|生成AI事例集 While many competitors are divided into those that do not have hardware and assume deployment to multiple companies from the start (Physical Intelligence, Skild AI, etc.) or those that provide their own hardware and AI as a single unit (Figure AI), Rhoda differs in that it clearly states as a business policy the sequence of proceeding to external expansion after verification through its own hardware.
SI001 Biped.News Rhoda AI Exits Stealth with $450 Million Series A to Build Robot Intelligence Platform | Biped.News Rhoda plans to use the capital to expand industrial pilots, grow its 100-plus person team, and begin licensing FutureVision to third-party robotics platforms targeting manufacturing and logistics.
SI002 Tech Funding News Khosla-backed Rhoda raises $450M at $1.7B valuation for video-trained AI — TFN The company's platform, called FutureVision, serves as the intelligence layer for the company's robotics systems and may eventually be licensed to partners building hardware or software for robotics platforms.
SI003 Rhoda AI Press Release | Rhoda AI Built on this architecture, 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.
SI004 Association for Advancing Automation (Automate.org) Rhoda AI's CEO on Building Mass Producing Humanoids Singh says the company has been focused on building a full-stack hardware/software solution since its earliest days. Rhoda attempted developing for existing systems, but found them insufficient for the kinds of industrial tasks it was looking to address.
SI005 SuperbCrew Rhoda AI Raises $450 Million In Series A Funding Round - SuperbCrew High valuations in the sector invite scrutiny; Rhoda must demonstrate real world ROI quickly to justify $1.7 billion amid competition from established players like Boston Dynamics and Tesla's Optimus.
SI006 The AI Insider Rhoda AI Exits Stealth with $450M Series A Funding Round to Scale Robitics Intelligence Platform FutureVision serves as the company's core intelligence platform and is expected to function as a foundation model that can be deployed across different robotic systems. Rhoda AI said it plans to license the technology to partners using a range of hardware and software platforms.
SI007 Simplify / Rhoda AI Research Member of Technical Staff @ Rhoda AI | Simplify We've raised over $450M and are investing aggressively in model research, infrastructure, hardware development, and manufacturing scale-up to make generalist robotics a reality.
SI008 Mayfield Fund Why We Backed Rhoda AI: The $30 Trillion Physical AI Market At 1,000 deployed units, Rhoda generates ~$100M in recurring annual revenue. At 10,000 units, it's a billion-dollar ARR business. The unit economics are as compelling for the customer as they are for Rhoda – because every robot deployed unlocks more output from the humans alongside it.
SI009 IndexBox Rhoda AI FutureVision Platform & $450M Funding at $1.7B Valuation - News and Statistics - IndexBox Nevertheless, analysts note that significant obstacles such as reliability, safety certification, and cost must still be overcome before general-purpose robots see widespread commercial use.
SE001 Inside DeepTech A comprehensive guide to Rhoda AI and its approach to robotics Rhoda AI is pioneering a shift in how robots learn, moving away from rigid programming to models informed by massive visual datasets.
SE002 WeTalkRobots Rhoda AI Advances Robotics with 'Direct Video-to-Action' Model Trained on Web-Scale Video The Direct Video-to-Action (DVA) model is a foundational robotics intelligence system developed by Rhoda AI. It redefines robot policies as video generation, utilizing hundreds of millions of openly accessible online videos for pre-training, similar to how large language models are trained on vast text data.
SE003 Rhoda AI Causal Video Models Are Data-Efficient Robot Policy Learners | Rhoda AI Direct Video-Action Model: A robot policy that translates predictions from a pre-trained causal video model into actions in a real-time closed loop, with the video model directly responsible for decision-making.
SE004 Gleam Direct Video Action Models — Gleam DVA has three interlocking pieces. Each has a distinct job, trained on different data at different scales.
SE005 Association for Advancing Automation Rhoda AI's CEO on Building Mass Producing Humanoids If we have to e-stop the robot, we don't have to worry about the robot collapsing and either damaging itself or worse injuring a human being
SE006 SuperbCrew Rhoda AI Raises $450 Million In Series A Funding Round - SuperbCrew Rhoda also develops its own robot hardware platform, rated for 25kg payload (40kg peak), with features like safety rated vision, wheel base brakes, high stiffness, and reliability for three years of continuous operation.
SE007 The AI Insider Rhoda AI Exits Stealth with $450M Series A Funding Round to Scale Robitics Intelligence Platform Rhoda AI said its Direct Video Action architecture integrates perception and control in a closed-loop system that allows robots to update actions continuously as conditions change, rather than relying on fixed plans that can fail in dynamic environments.
SE008 Simplify Research Member of Technical Staff @ Rhoda AI | Simplify Design and implement scalable curation pipelines for web-scale video pretraining data: ingestion, deduplication, quality filtering, and content classification across internet-scale video corpora
SU001 RoboHorizon Rhoda AI Unveils Video-Trained Robots, Nabs $450M at $1.7B Valuation Rhoda AI’s platform, dubbed FutureVision, aims to solve a long-standing problem in robotics: creating machines that can adapt to messy, real-world environments instead of being confined to rigidly programmed tasks.
SU002 YesPress Rhoda AI - Robots That Learn By Watching the World A component-processing workflow completed in under two minutes per cycle, autonomously - reportedly exceeding customer KPIs.
SU003 Rhoda AI Causal Video Models Are Data-Efficient Robot Policy Learners 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.
SU004 Association for Advancing Automation Rhoda AI's CEO on Building Mass Producing Humanoids The gap between the hype and the reality is is is quite substantial. We just did not want to be yet another company with just demos.
SU005 SuperbCrew Rhoda AI Raises $450 Million In Series A Funding Round High valuations in the sector invite scrutiny; Rhoda must demonstrate real world ROI quickly to justify $1.7 billion amid competition from established players like Boston Dynamics and Tesla’s Optimus.
SU006 The AI Insider Rhoda AI Exits Stealth with $450M Series A Funding Round to Scale Robitics Intelligence Platform In one manufacturing evaluation, Rhoda reported that its system autonomously completed a component-processing workflow in under two minutes per cycle while meeting production performance benchmarks.
SU007 Simplify / Rhoda AI Research Member of Technical Staff @ Rhoda AI We own the full robotics stack from high-performance hardware and robot systems to the infrastructure and state-of-the-art foundation world models that control our robots.
SU008 Mayfield Why We Backed Rhoda AI: The $30 Trillion Physical AI Market At 1,000 deployed units, Rhoda generates ~$100M in recurring annual revenue. At 10,000 units, it’s a billion-dollar ARR business.
SR001 Kazumichi Moriyama Rhoda AI Announces 'DVA' Model That Learns Action Policies Directly from Video, Enabling Multi-Step Tasks and Raising $450 Million While it has shown initial results in strong action data efficiency and long-term visual memory, it also acknowledges challenges such as achieving robust planning and reasoning, self-improvement through interaction, and long-term mobile manipulation with high-degree-of-freedom precision control.
SR002 Udit Rhoda AI exits stealth with $450M to build video-trained robots
SR003 Rhoda AI Causal Video Models Are Data-Efficient Robot Policy Learners At the same time, major challenges remain ahead: enabling robust planning and reasoning, building systems that can improve themselves through interaction, and unlocking long-horizon mobile manipulation and high-degree-of-freedom dexterous control.
SR004 Association for Advancing Automation Rhoda AI's CEO on Building Mass Producing Humanoids Singh confirms, however, that Rhoda has been testing its proprietary hardware with customers.
SR005 SuperbCrew Rhoda AI Raises $450 Million In Series A Funding Round High valuations in the sector invite scrutiny; Rhoda must demonstrate real world ROI quickly to justify $1.7 billion amid competition from established players like Boston Dynamics and Tesla’s Optimus.
SR006 The AI Insider Rhoda AI Exits Stealth with $450M Series A Funding Round to Scale Robitics Intelligence Platform The company said the funding will support research and engineering, expansion of industrial deployments and growth of its technical team.
SR007 Rhoda AI via Simplify Research Member of Technical Staff @ Rhoda AI This team owns web-scale video data curation, annotation pipelines, and evaluation methodology — directly determining the quality of the video pretraining distribution and how clearly we can measure model progress.
SR008 Mayfield Why We Backed Rhoda AI: The $30 Trillion Physical AI Market At 1,000 deployed units, Rhoda generates ~$100M in recurring annual revenue. At 10,000 units, it’s a billion-dollar ARR business.
SR009 AmericaBots Rhoda AI exits stealth with $450M to train robots from video Most critically, the company hasn’t addressed the liability and safety-validation frameworks required for autonomous manufacturing systems. Industrial customers demand determinism and fault-tolerance; a stochastic video-generation model may face regulatory and insurance barriers that venture capital can’t overcome.
SV001 Creati.ai Rhoda AI Raises $450 Million at $1.7 Billion Valuation to Train Robots Using Internet Videos Palo Alto-based Rhoda AI has officially exited its 18-month stealth phase to address this exact challenge, announcing a monumental $450 million Series A funding round. This massive capital injection catapults the company to a post-money valuation of $1.7 billion, signaling immense market confidence in its paradigm-shifting approach to robotic intelligence.
SV002 futureTEKnow Rhoda AI robot intelligence hits $1.7B Key unknowns for operators and investors include: Total cost of ownership compared with traditional automation cells and human labor. Reliability over multi-shift operations, including failure modes when predictions are wrong or sensors degrade. ... independently verified performance data across multiple sites and verticals has yet to emerge publicly.
SV003 Rhoda AI Press Release | Rhoda AI The $450 million Series A will support continued research and engineering investment, expansion of industrial deployments and customer pilots, and growth of Rhoda’s multidisciplinary team spanning generative AI, computer vision, and robotics.
SV004 Business Wire Rhoda AI Exits Stealth with $450 Million Series A to Bring Robots Out of the Lab and Into the Real World In a recent high-volume manufacturing evaluation, Rhoda completed a component-processing workflow in under two minutes per cycle without human intervention, exceeding customer KPIs.
SV005 Association for Advancing Automation Industry Insights: Rhoda AI's CEO on Building Mass Producing Humanoids But the reality is, despite all the billions that have been raised, despite all the cool videos that have been published, I actually can't think of a single example of a generalist autonomous robot that's doing useful work in the field today. The gap between the hype and the reality is is is quite substantial.
SV006 SuperbCrew Rhoda AI Raises $450 Million In Series A Funding Round - SuperbCrew High valuations in the sector invite scrutiny; Rhoda must demonstrate real world ROI quickly to justify $1.7 billion amid competition from established players like Boston Dynamics and Tesla’s Optimus. Technical hurdles include ensuring model safety in high stakes industrial environments and addressing potential data biases from web videos.
SV007 The AI Insider Rhoda AI Exits Stealth with $450M Series A Funding Round to Scale Robitics Intelligence Platform The company said models pretrained on large-scale video data learn underlying motion patterns and physical interactions, enabling robots to acquire new manipulation tasks with relatively small amounts of teleoperation data, in some cases requiring as little as about ten hours of robot training.
SV008 Simplify Research Member of Technical Staff @ Rhoda AI | Simplify We've raised over $450M and are investing aggressively in model research, infrastructure, hardware development, and manufacturing scale-up to make generalist robotics a reality.
SV009 Mayfield Fund Why We Backed Rhoda AI: The $30 Trillion Physical AI Market At 1,000 deployed units, Rhoda generates ~$100M in recurring annual revenue. At 10,000 units, it’s a billion-dollar ARR business. The unit economics are as compelling for the customer as they are for Rhoda – because every robot deployed unlocks more output from the humans alongside it.