Multi-agent Reinforcement Learning-based Twin-vehicle Fair Cooperative Driving in Dynamic Highway Scenarios

Highway is an important scenario for autonomous driving application because of its clear rules and little social intervention. In this scenario, cooperative driving of the unmanned vehicles is also a key technology. To achieve a simpler system architecture and lighter computation than rules-based cooperative driving methods, a multi-agent reinforcement learning-based twin-vehicle cooperative driving method is proposed in this paper. This work implements the generalization adaptation of reinforcement learning method in high dynamic highway scenarios. Besides, it pays equal attention to the autonomy of each one and their cooperation through a fair cooperation algorithm, realizing the independent lane changing and overtaking in heavy traffic, while keeping a fixed formation in loose traffic. Thus, the twin-vehicle can speed up while avoiding the interference of rigid structure on traffic. Experiments in a variety of highway scenarios verify the cooperative performance, also further increase the possibility of creating a harmonious driving environment.