Efficient multi-agent reinforcement learning through automated supervision