Dream Labs raises $90M to help build more intelligent robots with simple video feeds

Dream Labs, a startup aiming to create more intelligent robots that can perform various tasks, has secured $90 million in seed funding for the launch of a public application programming interface (API) providing access to its universal robot foundation model. The funding round was co-led by B Capital and Kindred Ventures and saw participation from other investors including A.Capital, Unusual Ventures, Wing VC, SV Angel, Scribble VC and angel investor Adrian Aoun.

Currently, industrial robots, numbering over five million globally, are highly specialized and perform a single task efficiently, but they lack the ability to learn new tasks or share acquired knowledge with other robots. The high costs of robotic hardware, which sell for just a few thousand dollars, require extensive programming to teach them specific jobs, which can take months. Moreover, this process must be repeated whenever new hardware is introduced.

With the launch of Dream-1, a versatile foundation model, Dream Labs aims to solve this issue. Dream-1 treats robotic skills as simple video prediction problems, learning from "human egocentric training data" like videos of people performing physical tasks. This approach allows Dream-1 to quickly adapt to new tasks, transferring intelligence across different robotic hardware. Developers can access Dream-1 through the Dream API, paying $3 per hour for the most powerful version and $0.75 per hour for a lightweight version.

Dream Labs is currently working with over 20 robotic hardware companies, optimizing the model for specific hardware to enhance performance. The company has ambitions to expand Dream-1's capabilities to support mobile robots with wheels and robots with humanlike hands. The funding will support the necessary compute power to achieve these goals.