Human-level performance in 3D multiplayer games with population-based reinforcement learning