Distributed iterative reinforcement learning predictive control of truck platoons

Addressing the issue of insufficient real-time computational capability of the centralized controller in solving multi-objective and multi-constraint nonlinear optimization problems for truck platoons, this paper proposes a synchronous distributed model predictive control strategy based on the Predecessor-Leader-Following communication topology. This approach transforms the global optimization problem of the platoon into local optimization problems for each truck, allowing all following trucks to solve their own optimization problems in parallel. Addressing the challenges of behavior prediction arising from the strong coupling characteristics of truck dynamics, a five-degree-of-freedom nonlinear dynamics model that captures both lateral and longitudinal coupling is developed to predict truck behavior. Additionally, a lane-keeping model is formulated to ensure that the longitudinal velocity of the trucks in the platoon matches that of the lead truck, while keeping the trucks within the designated lane. To reduce computational burden, a distributed iterative reinforcement learning predictive control scheme based on actor-critic networks is introduced. Co-simulation results using Matlab/Simulink and TruckSim demonstrate that the proposed strategy ensures both longitudinal velocity tracking and lateral lane-keeping performance, while providing better computational efficiency than conventional nonlinear model predictive control algorithms.