智能网联汽车基于逆强化学习的轨迹规划优化机制研究

Trajectory planning is one of the most significant technologies of autonomous connected vehicles. However, there are some problems in existing trajectory planning strategy, for example, weak real-time ability, difficult to calibrate weighting coefficients of optimization objectives and the poor interpretability for direct imitation learning method in the trajectory planning strategy. Therefore, an inverse reinforcement learning (IRL) method was proposed based on the maximum entropy principle in this paper. Learning the underlying optimization mechanism of driving trajectories from experienced drivers, the planning of lane-changing expert trajectories was achieved aligning with the human driving experience, laying a theoretical foundation for solving the real-time and interpretability problems of trajectory planning methods. Finally, taking general risk scenarios and high-risk scenarios as application cases respectively, the feasibility and effectiveness of the proposed trajectory planning method were validated through Matlab/Simulink simulations.