Skil AI Cracks Robot Football with Simulated Self-Play Training
Skil AI has made significant strides in robotics with its S1 model, which has been trained to play football using self-play training method within NVIDIA Isaac Sim. The model's objective was simple: score goals, without any tailored rewards or human hand-holding.
The result is a robot that can now play football against both humans and other robots in real-world scenarios, after being compressed into just a few weeks of wall-clock time from 140 years' worth of simulated experience. This is made possible by modern GPU clusters, which can run thousands of parallel simulated environments simultaneously.
The S1 model's versatility extends beyond football, as it can perform complex manipulation tasks lasting up to 10 minutes with just a single video demonstration. Skil AI's broader ambitions lie in its development philosophy, drawing parallels to large language models that train on extensive data from human videos and physics simulations to build general-purpose motor intelligence.
NVIDIA's simulation infrastructure plays a critical role in this strategy, providing the physics engine and rendering pipeline for high-fidelity robotic training at scale. Skil AI has already achieved significant commercial success, reaching a $100 million annual recurring revenue run rate shortly after its first deployment in 2026, with clients operating across more than 60 companies.
The company has also secured $1.4 billion in funding from investors including SoftBank and NVIDIA, achieving a valuation exceeding $14 billion.