Skild AI Taps NVIDIA Physical AI to Teach Robots New Tasks From a Single Video
nvidia ·
Skild AI has launched its S1 robot foundation model, which uses NVIDIA physical AI technologies to help robots learn new tasks from a single video demonstration. The model is designed to learn previously unseen, long-horizon tasks without updating its weights or undergoing task-specific post-training, using a technique called in-context learning. Skild built and researched S1 on NVIDIA AI infrastructure as part of a collaboration covering synthetic data generation, model training, simulation and real-world deployment. The company reached a $100 million annual revenue run rate 10 months after its first commercial deployment and has built more than 60 deployment partnerships in manufacturing, logistics, inspection, security and food preparation. S1 accepts an operator-recorded video as a prompt, interprets the demonstrated intent, objects and sequence, and maps them to robot actions without retraining, often for tasks outside its pretraining data. It can perform unfamiliar tasks lasting up to 10 minutes, such as plant potting, pancake making, pour-over coffee brewing and kit assembly. In tests, S1 succeeded about 66% of the time at each step of new multistep tasks, compared with 9% for a similar AI system, and one short video example was estimated to be as useful as roughly 380 hands-on examples. Skild, NVIDIA and Foxconn are deploying the Skild Brain on dual-arm manipulators for high-precision assembly of NVIDIA Blackwell systems, while NVIDIA tools including Cosmos, Omniverse, Isaac Sim, Isaac Lab and Newton support the development cycle.
nvidia