Skild AI Uses NVIDIA Technology to Teach Robots Tasks From One Video

Skild AI introduced its S1 robot foundation model, which enables robots to perform previously unseen multistep tasks from a single video demonstration. The model uses the video as an instruction and does not require weight updates or task-specific post-training.

S1 can perform unfamiliar tasks lasting up to 10 minutes, including plant potting, pancake making, pour-over coffee brewing and kit assembly. It interprets the demonstrated intent, objects and sequence, then transfers them into actions for the robot in front of it.

In a plant-potting test, the transition from recording the demonstration to autonomous execution on hardware took 11 minutes. The model can also respond when objects move, recover from errors and combine skills in sequences that were not explicitly programmed.

S1 test results

On new multistep tasks, S1 achieved about 66% success at each step, compared with 9% for a similar system. Skild AI estimates that one short video can be as useful as roughly 380 hands-on examples; collecting that amount manually could take 50–100 hours.

Skild built and researched the model on NVIDIA infrastructure. The collaboration covers synthetic data generation, model training, simulation and deployment of physical AI in real-world environments.

Skild AI, NVIDIA and Foxconn are deploying Skild Brain on dual-arm manipulators for high-precision assembly of NVIDIA Blackwell systems. In one demonstrated workflow, a robot installs a busbar and limit block, fastens 16 screws and adapts to disturbances during the task.

Skild uses NVIDIA Cosmos, Omniverse, Isaac Sim, Isaac Lab, TensorRT and other tools in development. The companies are also jointly developing GPU-accelerated simulation solvers for modeling how robots contact, grip and manipulate solid objects.