Event-Triggered Learning for Intelligent Connected Vehicle Platoon Optimization Within a Differential Game Framework

This paper proposes an event-triggered learning scheme to approximate distributed Nash equilibrium and solve the platoon control problem of heterogeneous multi-vehicle systems with unknown dynamics. The platoon control problem is modeled as a multi-agent differential graphical game. The off-policy Bellman equation is theoretically derived, and its convergence and optimality are illustrated. A data-driven reinforcement learning algorithm is thus designed to approximately solve multiple coupled Hamilton-Jacobi equations, and is implemented through a single-critic neural network structure. Then, a dynamic auxiliary variable that leverages static triggering information is constructed to promote distributed dynamic triggering. Finally, the effectiveness of the proposed learning scheme is verified on an intelligent connected vehicle platoon system.