Robust Longitudinal Control for Vehicular Platoons Using Deep Reinforcement Learning

In the last few years, researchers have applied ML strategies in the context of cooperative transportation to increase safety and efficiency. The RL paradigm has been successfully employed in the longitudinal spacing control of CACC systems, but ensuring disturbance rejection remains a challenge. Characteristics such as uncertain parameters in the formation and external interference may imposed constant errors, preventing the agents from reaching the desired formation when traveling at cruising speed. Therefore, this paper proposes a new training model for platooning, such that the acceleration command of each agent provides a stable formation, according to the nonlinear control formalism. The design methodology is general enough to be applied across multiple topologies. Also, the outputs of the controllers design by ML have a term with integral action capabilities, enabling corrective actions to withstand uncertainties that are constant. The effectiveness of our proposal has been illustrated with experiments using different network topologies, uncertain parameters, and external forces. Comparative analyses were conducted against the state-of-the-art, whose findings offer new insights concerning the generalization and robustness of using () in the control of autonomous platoons.