Text this: Multi-agent reinforcement learning with layered autonomy and collaboration for enhanced collaborative confrontation