Skild AI Unveils 'S1', a Physical AI Model Powered by NVIDIA, Enabling Robots to Learn Long-Horizon Tasks from a Single Video
Key point
Skild AI has released 'S1', a foundation model that leverages NVIDIA infrastructure to enable robots to perform new long-horizon tasks using only a single video.
Details
Skild AI has unveiled S1, a robotic foundation model built on NVIDIA's physical AI infrastructure. S1 employs an in-context learning approach where, without weight updates or task-specific post-training, the robot interprets intent, objects, and sequence from a single video provided by an operator as a prompt and maps it to actions.
Learning Method and Performance
S1 can perform unfamiliar tasks not present in its pre-training data and handles long-horizon tasks consisting of dozens of steps lasting up to 10 minutes. In a plant potting test, it took 11 minutes from video recording to autonomous hardware execution, with the ability to adjust and recover from errors during object movement.
Benchmark results show that for new multi-step tasks, S1's step-wise success rate is approximately 66%, an improvement of more than 7x compared to similar AI systems (9%). A single short video is estimated to have the same effect as approximately 380 actual training examples.
Utilization of NVIDIA Technology Stack
Skild AI collaborated with NVIDIA throughout the entire process, from data generation to deployment. Key technologies utilized include:
- NVIDIA Cosmos: Diversifying training data and converting videos into structured descriptions
- Isaac Lab & Newton: Accurately modeling physical parameters to reduce the sim-to-real gap
- Omniverse & Isaac Sim: Testing edge cases and verifying behaviors in physics-based virtual environments
- TensorRT SDK: Supporting fast robot response through inference optimization
Business Status and Use Cases
Skild AI achieved an annual revenue run rate of $100 million within 10 months of its first commercial deployment and has signed over 60 partnerships in fields such as manufacturing, logistics, and security. Notably, in collaboration with Foxconn, Skild Brain was deployed for high-precision assembly tasks on NVIDIA Blackwell systems, performing complex multi-step operations such as busbar installation and screw fastening.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.