Event
Distributional Boundary Analysis of Zero Goals in G1 Shooting vs. Goalkeeping Adversarial Matchups
1 reports · 1 sources
A G1 humanoid robot striker trained with motion-prior reinforcement learning scored on all 60 penalty kicks, and a goalkeeper using historical encoding could save 85% of shots; however, when these two models were placed into the official adversarial scenario, 20 matches produced zero goals. A related analysis article used frame-by-frame data to identify the two-layer distributional boundary mechanism that causes this gap.
Report timeline
- Humanoid Robot Shooting and Goalkeeping Control Based on Motion-Prior Reinforcement Learning
A G1 shooter based on motion-prior reinforcement learning scored on all 60 penalty kicks, and a goalkeeper using historical encoding saved 85% of shots. However, when the two models were placed in the official adversarial scenario, they scored zero goals across 20 matches; the article uses frame-by-frame data to identify a two-layer distribution-boundary mechanism.
Guyuehome ·