Vehicle Driving Longitudinal Control Based on Double Deep Q Network

• Published in: International Conference on Measuring Technology and Mechatronics Automation (Print) pp. 273 - 275
• Main Authors: Zhang, Chi, Zhang, Xin, Ma, Peng, Dai, Sheng, Lu, Ying, Jiang, Li
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01.01.2022
• Subjects: Buildings; Deep Reinforcement Learning; Driving longitudinal control; Markov decision process; Process control; Sockets; Training; Uncertainty; Vehicle driving; Virtual environments
• ISSN: 2157-1481
• DOI: 10.1109/ICMTMA54903.2022.00059

Aiming at the problem that the adaptive cruise control system fails to follow the car due to the extreme motion of the preceding car during the following car following control, based on the analysis of the motion characteristics of the vehicle, a reward function that can consider the motion characteristics of the preceding car is proposed. The car-following problem is constructed as a Markov decision process under a certain reward function, deep reinforcement learning algorithms are brought in to solve the problem, and the Double Deep Q Network is used to establish a car-following strategy. The movement of the front and rear cars is regarded as a state change process. Uniformly accelerate linear motion, establish relational equations to solve, and train the car-following network with the solution as a reward, and conduct interactive iterative learning through the environment established with unity. Finally, the control strategy is tested based on real driving data.