DQN-Based Automatic Emergency Collision Avoidance Control Considering Driver Style

In this paper, a deep Q-network (DQN) longitudinal collision avoidance control method is proposed to achieve excellent safety, while improve the driving experience by considering driver style. In terms of network structure, the driver style information is concatenated with image and motion information between convolutional and fully connected neural networks, achieving the fusion of information from different dimensions. In addition, the ideal safe distance is designed into the reward function to achieve end-to-end longitudinal collision avoidance control considering driver style. A simulation experimental environment is established based on Carla to test the control effects. The simulation results show that the control strategy of proposed method has a considerable effect on handling longitudinal collision avoidance problems. The longitudinal collision avoidance control of the vehicle exhibits distinct variations under different driver styles, and the longitudinal collision avoidance process can meet the psychological expectation of drivers.