Skild AI unveiled S1, a robot foundation model that can learn new tasks from a single video without retraining or task-specific fine-tuning, using in-context learning.

S1 can handle unfamiliar tasks up to 10 minutes long with dozens of manipulation steps, adapting when objects shift and recovering from mistakes, with demonstrations including potting plants, making pancakes, brewing pour-over coffee, and assembling kits.

The model interprets demonstrated intent and sequences from a video prompt and maps them into robot actions, handling tasks not covered by its pretraining data and performing tasks up to ten minutes long.

NVIDIA is a notable backer, and Skild has partnerships with major robotics manufacturers ABB and Universal Robots, underscoring investor confidence in a rapidly expanding market.

The technology is already being deployed with Foxconn on dual-arm manipulators for high-precision assembly of NVIDIA Blackwell systems, enabling a workflow that includes installing components, fastening screws, and adapting to disturbances.

NVIDIA technologies underpin the development and deployment pipeline, including simulation with Omniverse and Isaac Sim, Cosmos open-world models for diverse data, Isaac Lab for reinforcement learning, and Newton-based physics to model contact and manipulation.

S1 relies on NVIDIA technologies such as Isaac Lab, Cosmos, Omniverse, Isaac Sim, and TensorRT to enable simulation, synthetic data generation, training, and fast inference on real robots.

The company collaborates with NVIDIA to leverage Cosmos technologies for translating video demonstrations into robotic actions and uses NVIDIA Isaac Sim for physics-based virtual environments and reinforcement learning to refine capabilities.

Earlier this year, Skild acquired Zebra Technologies’ robotics division to expand into warehouse and logistics applications and markets its AI as a robot-agnostic brain capable of operating across multiple robot types and environments.

Skild AI reports reaching a $100 million annual revenue run rate ten months after its first commercial deployment and has formed over 60 partnerships across manufacturing, logistics, and food preparation.

Commercial traction includes over 60 deployment partnerships across manufacturing, logistics, inspection, security, and food preparation; on Foxconn’s lines, Skild Brain on dual-arm manipulators is used for high-precision assembly of NVIDIA Blackwell systems, handling tasks like busbar installation and screw fastening with disturbance recovery.

NVIDIA details a collaboration with Skild AI showing how Skild built its S1 robotic foundation model on NVIDIA AI infrastructure, achieving a $100 million annual revenue run rate within 10 months of first commercial deployment.