Vehicle Trajectory Prediction Based on Attention Reinforcement Learning and Hidden Markov Decision Network

The field of intelligent vehicle research is rapidly advancing, driven by the integration of reinforcement learning techniques. This study explores the use of virtual vehicles in simulated environments to iteratively navigate multiple routes under diverse traffic conditions, aiming to identify optimal driving paths. Accurate trajectory prediction of surrounding vehicles is essential for enhancing the operational efficiency and safety of intelligent vehicles, while also contributing to the overall improvement of traffic systems. It not only increases travel efficiency but also significantly reduces the risk of traffic accidents. This paper introduces a novel approach that employs a reinforcement learning framework combined with an Attention-mechanism Twin Delayed Deep Deterministic Hidden Markov Network (ATTD-HMMN) for collaborative decision-making. The method differentiates between driver-centric and external observational perspectives, integrating them with spatial and modal travel features to analyze both physical and state-level vehicle information. The model is evaluated on two publicly available datasets, NGSIM and HighD. Experimental results show that the proposed algorithm outperforms traditional model-based methods in predicting vehicle maneuvers and future trajectories, confirming the effectiveness and robustness of the proposed approach.