A San Francisco startup founded by the former leads of Nvidia's GR00T robotics research team has raised a $90 million seed round to ship what it calls the first publicly available API for a robot foundation model. Dream Labs, founded in 2026, announced the round on Monday led by B Capital with Kindred Ventures, joined by A.Capital, Unusual Ventures, SV Angel, Wing VC, Scribble VC and angel investor Adrian Aoun.
The flagship product is Dream-1, a model that takes a robot's camera feed and a plain-English instruction and outputs motor commands directly. Alongside the funding, the company launched Dream API, which lets developers connect any robot with a gripper and have it carry out tasks without writing task-specific code. A browser-based Robot Playground lets anyone type commands to a physical robot in the company's San Francisco office and watch it respond live and unedited.
The pitch targets a well-known bottleneck. Factories worldwide run a record 5 million industrial robots, yet nearly every one is programmed for a single task, and what one machine learns does not transfer to another. Low-cost arms now sell for between roughly $2,000 and $10,000, but they ship with no built-in intelligence, and teaching one robot one task can take specialist engineers months.
Dream-1 was trained on video of people performing everyday tasks rather than on demonstrations recorded from a specific robot, which the company says is what lets skills transfer across hardware. Out of the box, the company claims the model can control any gripper-equipped robot zero-shot, and it currently handles tabletop tasks such as picking and placing objects and folding laundry, with wheeled robots and dexterous hands next in line. For its more than 20 hardware partners, Dream Labs refines the model per robot.
"The hardware is ready. What has been missing is a brain any robot can use," co-founder and CEO Joel Jang said in the announcement, comparing the experience to getting into a rental car and driving off without relearning the skill. B Capital general partner Daisy Cai credited the team with "some of the defining work in robot learning at Nvidia," while MakerMods CEO Ryan Chan said his $2,000 bimanual arms were attempting tasks the company never programmed within minutes of running Dream-1.
Beyond the API, the company sells Fine-tuning as a Service, turning small batches of demonstration data into tailored models within hours for unsupported hardware. The new funding will go primarily toward compute as Dream Labs expands the range of tasks and robots Dream-1 can handle.
The robotics foundation model field is getting crowded — Physical Intelligence, Skild AI and Nvidia's own GR00T effort are all chasing generalist robot brains — but few have exposed their model through a public, self-serve API this early. All capability claims above are company-reported and have not been independently verified; whether video-learned skills hold up outside tabletop settings is the question customers will now test with their own API keys.
Comments (0)
Log in to join the discussion
Log InNo comments yet