Signal Skild AI Launches S1 Robot Foundation Model with One-Shot Video Learning

Summary

Robotics startup Skild AI unveiled its robot foundation model S1 on August 25, 2026. The model takes a single first-person human demonstration video as its only instruction and carries out the task immediately, with no fine-tuning or additional post-training. On tasks absent from its pretraining, spanning up to ten minutes and dozens of manipulation steps, it achieved a 66% success rate, compared with 9% for an equivalent language-prompted policy. Skild demonstrated this performance on four previously unseen tasks: potting a plant, cooking pancakes, making pour-over coffee, and assembling a kit. The announcement came without released weights, an API, or an accompanying paper. The launch points to rapid progress in transferring human physical skills to robots purely through video observation.