How would you design a multi-agent reinforcement learning system for collaborative autonomous behavior at Anduril Industries?