Skild AI's S1 learns robot tasks from one video, hits $100M revenue run rate

NVIDIA Blog · rss · 2026-09-11

Skild AI launched its S1 robot foundation model in collaboration with NVIDIA, designed to let robots adapt to changing factory work without reprogramming.

Core capability:

Numbers: 66% per-step success on new multistep tasks vs. 9% for a comparable system (7x+); one short video ≈ 380 hands-on training examples, which would take 50–100 hours to collect manually.

Commercial traction: $100M annual revenue run rate 10 months after first deployment, 60+ deployment partnerships across manufacturing, logistics, inspection, security, and food prep. Skild, NVIDIA, and Foxconn are deploying Skild Brain on dual-arm manipulators for high-precision Blackwell system assembly.

Stack: NVIDIA Cosmos for synthetic data and video structuring, Isaac Sim/Omniverse for simulation and edge-case testing, Isaac Lab (Newton physics engine) for RL to close the sim2real gap, TensorRT for inference latency. Joint GPU-accelerated contact/grasp solvers will ship in Newton for all developers.

Original post →

More from Embodied

Embodied channel →