Under the mixed traffic flow environment, autonomous vehicles (AVs) may conflict with manual-driven vehicles (MVs), particularly at intersections. For this purpose, the future trajectories of conflicting vehicles present countless possibilities, which significantly reduces the safety of AV when navigating through intersections. To address this issue, this paper presents a reinforcement learning (RL) based two-dimensional optimization algorithm, aiming to enhance AV efficiency, safety, and comfort to deal with possible conflicts in the intersection area. We present the MVs trajectories in the form of a probability distribution range, supposing the possible MVs' future trajectories. Using the MVs trajectories uncertainty, a Markov decision process is used to develop a motion model for AVs, and RL is applied to find the optimal strategy. To verify the proposed method, we used the dataset of Gaoke East Road-Tanglu Road in Pudong New Area, Shanghai, and compared two parameters such as safety and efficiency. During the simulation process, results show that the proposed method's safety is better than optimal control methods based on shortest time (OC-ST), artificial potential field methods (APF), and game theory (GT). In the preservative of efficiency, the proposed method is performing better than APF and GT and has less efficiency during the comparative analysis with OC-ST.
Two-Dimensional Trajectory Optimization for Autonomous Vehicles at Intersection
06.12.2024
405940 byte
Aufsatz (Konferenz)
Elektronische Ressource
Englisch
Taylor & Francis Verlag | 2023
|