Skild AI publishes S1 in-context robotics research
Skild AI, Inc. →Skild AI published research describing S1, its robotic foundation model designed to interpret a task from a video demonstration. The company reported that the same model weights could perform both familiar tasks in new configurations and previously unseen tasks without task-specific fine-tuning. Demonstrations included plant potting, pancake preparation, coffee brewing and kit assembly, with sequences lasting up to ten minutes. In one plant-potting example, Skild reported eleven minutes from the start of recording a human demonstration to autonomous execution. S1 was built using NVIDIA AI infrastructure. The research discusses learning across diverse tasks, adapting to disturbances and recovering from errors. These are company-reported research results, rather than a guarantee of performance across every deployment.