关键词: 拟人化决策；驾驶行为建模；逆强化学习；轨迹预测；轨迹规划

Abstract: To bridge the gap between decision-making and planning behaviors of autonomous vehicles and human drivers, this paper proposes a personalized decision-making and planning approach based on maximum entropy inverse reinforcement learning (MaxEnt IRL). First, a convolutional-pooling long short-term memory (LSTM) neural network model is used to capture the interaction relationships of surrounding vehicles and predict their trajectories. Second, during the modeling of driving behavior, continuous human behaviors are discretized to reduce the computational complexity of MaxEnt IRL, and a personalized reward function is introduced to reflect the preferences and decision-making processes of individual drivers. Subsequently, a quintic polynomial planning method is employed to solve the trajectory. Finally, the effectiveness of the proposed method is validated in a simulation environment. Experimental results demonstrate that, compared with conventional IRL algorithms, the proposed personalized MaxEnt IRL algorithm significantly reduces the error between planned trajectories and expert trajectories, with the root mean squared error (RMSE) decreasing from 3 m to 0.8 m, while notably enhancing passenger ride comfort.

Key words: human-like decision-making; driving behavior modeling; inverse reinforcement learning; trajectory predicting; trajectory planning