NVIDIA-Powered Robots Learn New Tasks with Skild AI's S1 Model

Skild AI has developed its S1 model, a robotic foundation powered by NVIDIA's technology. The AI system enables robots to learn new tasks from single video demonstrations, making them more adaptable in various industries such as manufacturing and logistics.

The key innovation behind S1 lies in its ability to use video prompts for 'in-context learning.' An operator records a task and feeds it to the model. From there, the robot interprets objects, intent, and sequence of actions, autonomously translating them into executable steps.

S1 can complete complex tasks lasting up to 10 minutes, often involving dozens of sequential actions. In internal tests, the model executed tasks successfully 66% of the time, an over sevenfold improvement compared to similar AI systems, which achieved a 9% success rate.

The S1 model builds on NVIDIA's broader strategic partnership with Skild, incorporating technologies like Isaac Lab and Cosmos for simulation, synthetic data generation, and scalable training. This collaboration aims to make robots more adaptable and efficient in real-world applications.