부산대학교 도서관

In recent years, self-driving car technology has made increasingly rapid progress with advances in deep learning and artificial intelligence.Intelligent perception of the surrounding environment and rational planning of driving routes are crucial for safe driving. Automated vehicles have the ability to predict the future trajectory of traffic participants around them, which helps them avoid accidents. Therefore, accurately predicting the future trajectories of traffic participants around self-driving vehicles is a key issue for reasonable route planning and safe driving of self-driving vehicles.After that, this paper designs a multi-objective trajectory evaluation function for the passability of trajectories and the proximity of driver behavior, and evaluates and filters the trajectory clusters in various aspects. And the preferential selection of obstacle avoidance trajectories under different obstacle positions is completed using the above evaluation function. The fitted parameters of the preferred trajectories are used to train the BP neural network. The trained deep learning model is applied to the trajectory planner to achieve the planning of obstacle avoidance trajectories under different vehicle speeds and different obstacle positions.