A Distributed Approach to Autonomous Intersection Management via Multi-Agent Reinforcement Learning