Abstract
In this paper, we discuss collision avoidance for Connected and Autonomous Vehicles (CAVs) on a highway. CAVs are clustered into coalitions each managed by a leader. Within a coalition, collision avoidance is addressed using a Monte Carlo Tree Search (MCTS)-based approach. We propose algorithms for collision avoidance across coalitions. After an initial assessment of the impact of a potential collision on an affected coalition, leaders cooperate to define action plans that are free of intra-coalition and inter-coalition conflicts. The algorithms were validated through extensive realistic simulations in a multi-agent-based traffic simulator. Experimental results discuss the reliability and scalability of the algorithms for coalitions of different sizes. Moreover, we present an analysis to select the optimal coalition size and the optimal number of coalitions given a total number of CAVs.
Demos
Scenario: Acceleration misbehavior: Misbehaving vehicle accelerates from behind three coalitions
Scenario: Acceleration misbehavior: Misbehaving vehicle accelerates from behind two coalitions.
Scenario: Full stop misbehavior: Misbehaving vehicle breaks in front of three coalitions.
Scenario: Full stop misbehavior: Misbehaving vehicle breaks in front of two coalitions.