Skild AI Taps NVIDIA Physical AI to Teach Robots New Tasks From a Single Video
NVIDIA, Thursday, September 10th, 2026
The S1 robot foundation model learns manufacturing tasks from one video, hitting 66 percent per-step success versus 9 percent.
Skild AI's S1 robot foundation model learns complex manufacturing tasks from a single video demonstration without retraining, reporting a 66 percent per-step success rate against 9 percent for comparable systems. It executes multi-step tasks running up to ten minutes.
The model was built on NVIDIA infrastructure including Isaac Lab, Cosmos and Omniverse.
The single-demonstration learning capability is what changes the economics of robotic automation, since the cost of teaching a new task has historically dominated deployment cost in variable manufacturing environments.