AI agents taught themselves to play distinct soccer positions without being told how humans play
retained text
Amanda Prorok and her team tested whether AI agents would develop role specialization on their own in a simulated soccer game. The agents could either all learn the same behavior or specialize into different roles. They eventually formed distinct positions, including attackers and a goalkeeper that stayed back and defended the goal. The system was never told that this was how humans play soccer. It learned that behavior through reinforcement learning. [https://www.youtube.com/watch?v=NZ3iKgyI3gc](https://www.youtube.com/watch?v=NZ3iKgyI3gc)
document fetched 2026-09-15 00:39 · canonical https://reddit.com/r/robotics/comments/1wg427e/ai_agents_taught_themselves_to_play_distinct