Skild AI's S1 model teaches robots new tasks from one video

Skild AI announced its S1 robot foundation model last week. The model takes on tasks it has not seen before from a single demonstration video, without updating its weights or running task-specific post-training. By the company's own figures, S1 succeeds at roughly 66 percent of steps on new multistep tasks, against 9 percent for a comparable system. Skild says one short video can be worth about 380 hand-collected training examples, which a person would need 50 to 100 hours to gather. The model was trained on NVIDIA infrastructure. Skild, NVIDIA and Foxconn are working together on dual-arm manipulators that assemble NVIDIA Blackwell systems. The company said it reached a $100 million annual revenue run rate ten months after its first commercial deployment, with more than 60 deployment partnerships.