Skild AI teaches a robot a new task from a single video
Pittsburgh-based Skild has announced a robot foundation model, S1, designed to learn previously unseen multistep tasks from a single video demonstration. In one experiment reported by the company, eleven minutes elapsed between recording a plant-potting demonstration and the robot performing the task autonomously. The model can also adjust when objects move, recover from errors, and combine skills in sequences it was not explicitly programmed to perform.
This is not only a lab demo: Skild, Nvidia and Foxconn are running the model on dual-arm robots at an Nvidia factory in Houston, assembling Blackwell GPU systems. S1 was built on Nvidia infrastructure (Cosmos, Isaac Lab, Isaac Sim, Omniverse), and Skild says it reached a $100 million annual revenue run rate ten months after its first commercial deployment. Amit Goel, a product director at Nvidia, names the bottleneck that remains: data and evaluation.