Skild AI Uses NVIDIA Physical AI for Robot Learning

Skild AI has introduced the S1 robot foundation model, built using NVIDIA physical AI technology, with the ambitious goal of teaching robots new tasks from a single video demonstration. The model is designed to address a persistent problem in robotics: environments like manufacturing floors, warehouses, and production lines change constantly, and most robots require significant reprogramming whenever tasks or layouts shift. The S1 model targets previously unseen, long-horizon tasks, meaning multi-step activities that unfold over time rather than simple repetitive motions. By learning from video, robots could acquire new skills without the extensive engineering effort traditionally required to hand-code each behavior. That reduction in reprogramming burden is the core promise of the announcement. The work fits into a broader push toward general-purpose robot foundation models that can generalize across different embodiments, environments, and task types. Rather than building a bespoke system for every deployment, companies could rely on a shared foundation model and adapt it quickly to new situations. NVIDIA's physical AI stack provides the simulation, training, and inference infrastructure that makes this kind of learning feasible at scale. If successful, single-video task learning could dramatically lower the cost of deploying robots in dynamic industrial settings. Factories could reconfigure lines without lengthy robot retraining, and warehouses could adapt to seasonal changes more easily. The S1 model is an early step, but it points toward a future where robots learn on the job much as humans do, by watching and imitating rather than being explicitly programmed.