Dream Labs, a San Francisco startup founded this year by former leaders of Nvidia's GR00T robotics research team, came out of stealth on Monday, Oct. 19, with $90 million in seed funding and a public API that lets almost any two-armed robot take plain-English orders.

Why it matters: Most robots are programmed for one job, and teaching a new task can take specialists months. Dream Labs says its model can run robots it has never seen, starting at $3 an hour.

The deal

  • B Capital and Kindred Ventures led the round, with A.Capital, Unusual Ventures, SV Angel, Wing VC, Scribble VC and angel investor Adrian Aoun joining, the company said.
  • The money goes mostly to computing power to expand the tasks and robots its model can handle, according to the announcement.

How it works

The flagship model, Dream-1, is trained on video of people doing everyday tasks. It predicts what happens next and turns a robot's camera feed plus a typed instruction into motor commands.

The company says it works out of the box on two-armed robots with parallel grippers, with no fine-tuning. For now it handles tabletop jobs like picking and placing objects and folding laundry, with wheeled robots and dexterous hands to follow.

What it costs

  • $3 an hour for the flagship model, or $2 an hour for a slightly less capable one, Forbes reported.
  • More than 20 hardware partners, with supported robots costing about $2,000 to $10,000.
  • A free browser "Robot Playground" lets anyone type commands to a robot in its San Francisco office and watch it respond live.

More on robot training data: Santa Clara startup pays 5,000 people to film chores for robots.