Physics-Informed Multi-Agent Reinforcement Learning for Distributed Multi-Robot Problems