Cooperative Collision Avoidance for Autonomous Vehicles Using Monte Carlo Tree Search

Autonomous vehicles require an effective cooperative action planning strategy in an emergency situation. Most action planning approaches for autonomous vehicles do not scale well with the number of vehicles. In this dissertation, we present COCOA (Cooperative Collision Avoidance), an efficient cooperative action planning algorithm for autonomous vehicles in colliding situations. In COCOA, autonomous vehicles drive together in coalition formations for information sharing and cooperation. When a coalition member detects a colliding situation with a misbehaving vehicle, all coalition members explicitly cooperate to find conflict-free action plans to avoid collisions with the misbehaving vehicle. COCOA employs a hierarchical decision-making approach where action planning is achieved at two levels: at the vehicle level and at the coalition level. In emergency scenarios involving multiple coalitions, COCOA employs a sequential and hierarchical decision-making approach. Leaders of the coalitions in a coalition sequence cooperate to finalize action plans for their coalition members that are free of inter-coalition conflicts. The COCOA algorithm is validated through extensive realistic simulations in a multi-agent-based traffic simulation system.