Skild Ai · 3× · naposledy 27. 8. 2026
52
Momentum
S1 is the flagship robotic Foundation Model from Skild AI, unveiled in August 2026, designed as an In-Context-Learner: it requires only a single video demonstration of a task to execute it without Fine-Tuning or retraining. The model can thereby handle novel, long-horizon tasks of up to 10 minutes that were never seen during pretraining – a first among robotic Foundation Models. S1 translates the video demonstration provided in the context window directly into robot actions and was trained on NVIDIA AI infrastructure. Specific information regarding price, licensing model, or certifications is not publicly available on the official product page.
Vývoj momenta
01.06.30.08.
Vlastnosti
| Deployment Model | In-context learning at inference time: video demonstration drives robot action without changing model weights (no fine-tuning/post-training) |
| Use Case Scope | Manipulation and long-horizon tasks (up to 10 min) such as making coffee, potting a plant, frying pancakes – based on a single video prompt |
| Integrations | Built on NVIDIA AI infrastructure (incl. Omniverse, Isaac Lab, Cosmos) for training and simulation |
| Platform | Robotics foundation model, omni-bodied: works across different robot embodiments (humanoids, quadrupeds, tabletop arms, mobile manipulators) |
| Release Date | August 2026 |