Hierarchical optimization for multi-criteria lane-changing strategy and decoupled trajectory planning of autonomous vehicles in heterogeneous traffic

Abstract

Lane-changing is a crucial component of autonomous vehicles. In mixed multi-vehicle scenarios, traditional lane-changing strategies may reduce traffic efficiency due to high competition. To address this issue, this study proposes a lane-changing strategy and trajectory planning for enhancing driving safety and traffic efficiency. First, optimize the traditional mixed traffic flow model and establish the lane-changing feasibility assessment model. Next, combine the double deep Q-network (DDQN) algorithm to formulate a lane-changing strategy. Second, decouple trajectory planning, using the fifth-degree polynomial curve to plan the path. Then dynamic programming is employed for speed planning and multi-objective functions are used to optimize the trajectory planning results. Simulation and prototype validation indicate that compared with traditional methods, the proposed lane-changing strategy and trajectory planning can efficiently generate trajectories for autonomous vehicles, enhancing overall traffic efficiency and safety.