DDPG-based controller of enhanced adaptive cruise control with lane-change assistance for an articulated vehicle

Abstract

Traditional adaptive cruise control (ACC) would choose to follow when encountering low-speed cars in front, but this may lead to low driving efficiency. This paper proposes a controller of enhanced adaptive cruise control with lane-change assistance (LCACC) for an articulated vehicle. A two-layer hierarchical control structure is adopted in this study. The upper determines high-level commands, while the bottom consisting of two modified deep deterministic policy gradient (DDPG) networks, controls the steering wheel and throttle/brake, respectively, according to the commands made by the upper layer. The vehicle’s lateral and longitudinal control are decoupled and controlled by two modified DDPG networks. Through proper design of the state and reward function, the actions of the articulated vehicle such as steering and acceleration/braking are closer to those of humans, thus ensuring ride comfort. Compared with traditional ACC, the LCACC proposed in this study increases the driving efficiency of the articulated vehicle by an average of 16% when there is a low-speed lead car.