Cooperative Adaptive Cruise Control based on Reinforcement Learning for heavyweight BEVs

Cooperative Adaptive Cruise Control (CACC) is an Advanced Driver Assistance System (ADAS) that provides an important aid to highway mobility for heavy duty vehicles. At the same time, the rise of connectivity and artificial intelligence applied to the Automotive field pushes to investigate different solutions that can bring new advantages compared to traditional control strategies. This study proposes an alternative approach for a CACC problem based on Deep Reinforcement Learning (DRL) using Twin Delayed Deep Deterministic Policy Gradient algorithm (TD3) for heavyweight Battery Electric Vehicles (BEVs). The main goals of the proposed solution are an increased comfort with respect to the reference driving cycle and, in particular, a reduction of energy consumption thanks to a small average gap between the two vehicles, thus decreasing the air drag coefficient of the follower vehicle and consequently leading to a minor aerodynamic drag effect. Moreover, three different types of spacing involving minimum Time Headway and Time to Collision has been investigated in order to study the safety guarantees of the algorithm, especially in critical and unexpected situations such as a hard braking. The control strategy has been evaluated with respect to the Leader vehicle results and has been also compared with a linear MPC baseline. The achieved outcomes show that the Ego vehicle is able to save energy consumption up to 19.8% improving comfort at the same time.