Rhoda AI

What it builds

Rhoda AI is a Palo Alto company developing robot foundation models and, less publicly, its own robot hardware for industrial manipulation. It ended roughly 18 months of stealth on 10 March 2026, announcing $450 million in funding — which the company labels a Series A — and FutureVision, its approach to robot control based on video-predictive control. Bloomberg reported the round valued the company at $1.7 billion, and Humanoids Daily covered the $1.7B valuation and model launch at the time. The company says it works with customers across automotive, manufacturing, logistics and e-commerce.

Technical approach

Rather than training mainly on teleoperated robot trajectories, as vision-language-action pipelines typically do, Rhoda pre-trains an autoregressive video model on internet video to build a prior on motion, physics and interaction, then post-trains on smaller robot datasets to map predicted frames onto embodiment-specific actions. The company calls the result a Direct Video Action model and says it runs closed-loop, updating behaviour as conditions change rather than executing an open-loop plan. Rhoda states that this pre-training lets the model pick up new tasks with as little as ten hours of teleoperation data, and that long-context conditioning allows single-shot execution of tasks such as pick-and-place from a human demonstration without retraining. The pitch is aimed squarely at the lab-to-real-world generalization gap that CEO Jagdeep Singh has described as the sector's central problem.

Traction and hardware

Public evidence of deployment is limited and mostly company-supplied. Rhoda says its systems have run autonomously in production settings, citing a high-volume manufacturing evaluation in which a component-processing workflow was completed in under two minutes per cycle without human intervention. Demonstrations published so far include unlatching and collapsing heavy-duty plastic totes and handling parts on an automotive assembly line. Hardware has not been formally announced. In a July 2026 interview, Singh said Rhoda has been testing proprietary hardware with customers and resisted the humanoid label, describing a mobile manipulator with seven-degree-of-freedom arms, dexterous hands and a wheeled base rather than legs — and claimed Rhoda could be the first company to deploy a manipulation-capable autonomous robot at scale.

People and backing

Singh, previously founder and CEO of QuantumScape and Infinera, is co-founder and CEO. Forbes, citing internal documents, named Stanford professor Gordon Wetzstein and former SoftBank Pepper engineer Vincent Clerc as founding team members when it first reported the company's $230 million raised in stealth. Eric Ryan Chan, formerly a generative model architect at World Labs, is chief science officer. Disclosed investors include Capricorn Investment Group, Khosla Ventures, Leitmotif, Matter Venture Partners, Mayfield, Premji Invest, Prelude Ventures, Temasek, Xora and John Doerr.

Where it sits

Rhoda occupies an unusual middle position: it competes on models with video- and world-model-centric labs, while intending to control its own stack down to the robot, as hardware-first entrants do. Its wheeled base places it closer to mobile-manipulator vendors than to bipedal humanoid programmes, despite early reporting framing it as a humanoid company.