收录:
摘要:
Overtaking decision control in unmanned driving is a current research hotspot. Overtaking operations can improve driving efficiency, but blindly overtaking may cause traffic accidents. A critical issue is how to perform overtaking decisions in a specific driverless scenario safely. In this paper, based on Deep Deterministic Policy Gradient (DDPG) deep reinforcement learning algorithm, this paper establishes an end-to-end overtaking decision control model with continuous action output. The model uses low-cost sensors to obtain input state information (such as vehicle speed, position, road angle, etc.) and directly output vehicle action information (steering, acceleration, braking). The results show that the model can achieve end-to-end unmanned overtaking by training and testing the vehicle to complete the overtaking task in the TORCS (The Open Racing Car Simulator) platform environment. Finally, our work compared with the Deep Q-learning Network (DQN) overtaking model with discrete action output, and the experimental results show that the DDPG overtaking model has a better overtaking decision control effect. © 2023 IEEE.
关键词:
通讯作者信息:
电子邮件地址:
来源 :
年份: 2023
页码: 680-685
语种: 英文
归属院系: