Cooperative Control of Heterogeneous Connected Vehicle Platoons: A Dynamic Event-Triggered Reinforcement Learning Approach

Abstract

With the increasing demand for connected and autonomous vehicle (CAV) platoons in the intelligent transportation system, optimizing communication resources while ensuring control performance has emerged as a critical challenge. To this end, this paper proposes a novel event-triggered reinforcement learning (ETRL) scheme for cooperative control of heterogeneous CAV platoons. The platooning control problem is first formulated as a multi-agent differential graphical game. Within this framework, each vehicle’s error dynamics and predefined performance functions depend solely on local and neighboring information. Then, a data-driven and off-policy integral reinforcement learning algorithm is developed to solve the coupled Hamilton-Jacobi equations which latently characterize the distributed optimal control policy. To further conserve communication resources, a dynamic event-triggered mechanism is introduced, and a single-critic network structure is used to implement the ETRL-based platooning control scheme. The stability and optimality of the vehicular platoon system are substantiated by strict theoretical analysis. Finally, through comprehensive simulations incorporating aggressive stop-and-go behaviors and periodic longitudinal disturbances, the effectiveness and superior performance of the proposed scheme are definitively validated against existing benchmark algorithms across multiple indicators.