Founded by former leaders of Nvidia’s GR00T project, Dream Labs is launching Dream-1, a model trained on human videos that can control robotic arms from different manufacturers without the need for retraining.
Dream Labs, a San Francisco-based start-up founded in 2026 by former heads of Nvidia’s GR00T robotics research programme, has announced a $90 million seed round led by B Capital alongside Kindred Ventures, with participation from A.Capital, Unusual Ventures, SV Angel, Wing VC, Scribble VC and Adrian Aoun. The bulk of the funds will be allocated to computing power.
One brain for different robots
Its flagship model, Dream-1, learns from videos of people performing everyday tasks: it anticipates what will happen next in the scene and deduces the robot’s action from this. Because it is not tied to a specific machine, it can immediately control a two-armed robot fitted with grippers, without the need for additional data. It currently handles tasks on a table, such as picking up and placing objects or folding laundry; mobile wheeled robots and dexterous hands are set to follow.
At the same time, the company is launching Dream API, which converts a camera feed and instructions in everyday English into motor commands. Dream Labs works with more than 20 partner manufacturers, whose robots cost between approximately $2,000 and $10,000, and is refining the model for each one. Its CEO, Joel Jang, compares the approach to driving a hire car: you figure out the controls, and experience does the rest.
Read the full article on the original source
Go to the source →