Supervised Reinforcement Learning for Human-like Adaptive Cruise Control
This paper proposes a supervised reinforcement leaming (SRL) algorithm for the Adaptive Cruise Control system (ACC) with human-like driving habit needs, which can be thought of as a dynamic programming problem with stochastic demands. In short, the human-like ACC problem can be deemed as the host vehicle adopts different control parameters (accelerations in the upper controller, brakes and throttles in the bottom controller) in the process of following or other driving situations according to the drivers behavior. We discrete the relative velocity as well as the relative distance to construct the two dimensional state, and map it to a one dimensional state space; discrete the acceleration to generate the action set; design additional velocity improvement shaping reward and distance improvement shaping reward to construct the supervisor. We apply the SRL algorithm to the human-like ACC problem in different scenarios. The results show the higher robustness of the SRL control policy in the human-like driving mode compared with other traditional control methods, and the control system can have sufficient control accuracy in both the velocity and the distance.
Zhaohui Hu Dongbin Zhao
国际会议
哈尔滨
英文
239-248
2010-08-01（万方平台首次上网日期，不代表论文的发表时间）