Training agents by having them compete against a diverse population of opponents at different skill levels to improve robustness.