Skild taught a humanoid soccer via self-play — 140 simulated years, no human demos
lukas_m_ziegler · x · 2026-09-29
Skild AI used self-play to teach a humanoid robot soccer, with no hand-crafted rewards and no human demonstrations for the advanced behaviors.
After 140 simulated years of playing against itself, the robot learned to dribble past defenders, shield the ball and tackle opponents, then the policy was transferred to a real robot.
The rationale: imitation learning caps robots at human capability, whereas self-play is how AlphaGo beat humans at Go — and Skild sees it as the path to physical AGI.
More from Embodied
- Wayve's autonomous driving goes live in London with Uber, lit up across the city — alexgkendall · 2026-09-29
- Tesla FSD Supervised approved in Croatia, its 8th European market — mitchdeg · 2026-09-29
- Idle time in demos teaches robots to do nothing — dev open-sources a LeRobot dataset editor — lhoestq · 2026-09-29
- Microagi hires Georgia Tech professor Animesh Garg plus 7 researchers via stealth acquisition — lukas_m_ziegler · 2026-09-29
- Recursive Harness Distillation lifts robot manipulation success from 37.3% to 64.0% — SeoulNatlUniv · 2026-09-29
- Tsinghua humanoid robot plays badminton with one policy from just 30 min of human motion data — ChongZzZhang · 2026-09-29