AIs which develop faking, tackling, blocking, kicking through self-play
Competitive Self-Play: Self-play allows agents to discover a wide range of skills and learn transferable policies
Competitive self-play for AI agents
Watching wind push over an AI who trained itself to walk is so frustrating without the payoff. . #MachineLearning.
2/ Competitive self-play reinforcement learning results from , make it clear complex behaviors learnable