A safe lane-changing strategy for autonomous vehicles based on deep Q-networks and prioritized experience replay

Autonomous vehicles (AVs) still face many safety issues in lane change scenarios in dense highway traffic. This paper proposes a dense reinforcement learning approach based on deep Q-Network (DQN) and prioritized experience replay (PER), aimed at enhancing the lane change safety of AVs in dense highway traffic. By developing safety constraints and designing safety-first multi-dimensional reward functions, the proposed approach significantly improves lane-change safety in dense highway traffic. Furthermore, the conducted numerical simulation experiments demonstrate the outstanding performance of our approach.