Bilateral Control Model for Autonomous Vehicles Based on Deep Reinforcement Learning

Traffic oscillations degrade efficiency, increase safety risks, and lead to excessive energy consumption. To address this, we propose the Bilateral Control Model with Deep Reinforcement Learning (BCM-DRL), integrating Deep Reinforcement Learning (DRL), specifically the Twin Delayed Deep Deterministic Policy Gradient (TD3) algorithm to suppress oscillations and enhance stability. Using the I-80 Next Generation Simulation (NGSIM) dataset, BCM-DRL is trained and evaluated against Bilateral Control Model (BCM) and Car-Following Model with Deep Reinforcement Learning (CFM-DRL). Simulation results show that BCM-DRL reduces the cumulative damping ratio by 75%, decreases fuel consumption by 21.6%, and achieves near-zero Time-to-Instability (TIT) values, significantly improving stability and efficiency. These findings validate BCM-DRL as an effective approach to mitigating traffic oscillations and optimizing vehicle control.