An Integrated Prediction and Trajectory Planning Method for Mixed Traffic Environments

Abstract

Among the countless challenges confronting autonomous driving, a primary impediment to its real-world deployment lies in achieving safe and efficient decision-making and planning within dynamic and mixed traffic environments. The key to addressing this issue lies in the development of a sophisticated framework for trajectory prediction and planning that can effectively cope with the challenges posed by mixed traffic environments. This paper proposes an integrated research method that unifies trajectory prediction and trajectory planning to address this challenge. Firstly, the proposed method employs a deep learning model, based on the standard encoder-decoder architecture, to predict the future trajectories of surrounding vehicles. Secondly, based on these prediction results, A*-based search algorithm is employed to plan a collision-free, optimal trajectory for the ego-vehicle on the given map. Finally, the proposed method was comprehensively validated through simulations on the large-scale, high-fidelity highD real-world dataset. The experimental result corroborates the accuracy of our approach.