Joint Prediction of Intention and Trajectory in Automated Driving Using Interactive Dynamic Bayesian Network

• Published in: IEEE transactions on vehicular technology Vol. 74; no. 8; pp. 11857 - 11867
• Main Authors: Hao, Jinbo, Guo, Fan, Zhao, Dezong, Chen, Yun, Song, Kang, Xie, Hui
• Format: Journal Article
• Language: English
• Published: New York IEEE 01.08.2025; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Automation; Autonomous vehicles; Bayes methods; Bayesian analysis; Data mining; Decision making; Deep learning; Driving; dynamic interaction; Hidden Markov models; intention prediction; Lane changing; Prediction algorithms; Predictive models; Reinforcement learning; Trajectory; trajectory prediction; Transformers; Transportation safety; Vehicle dynamics; Vehicles
• ISSN: 0018-9545; 1939-9359
• DOI: 10.1109/TVT.2025.3557184

The prediction of surrounding vehicles' lane-changing intentions and trajectories is critical to the decision-making of autonomous vehicles. However, the precise trajectory prediction is challenging because of uncertainties in human intentions and complexities in vehicle interaction. To address these challenges, a new framework for joint prediction of vehicle intention and trajectory is proposed in this paper. The Interactive Dynamic Bayesian Network (IDBN) and the Transformer are used to establish the connections among vehicle interaction, lane-changing intentions and trajectories. Initially, the intentions of the surrounding vehicle and subject vehicle are modeled as nodes within the IDBN. The IDBN simulates the interaction of intentions and predicts the lane-changing intention of the surrounding vehicle. Subsequently, the lane-changing intention probabilities are fused by the Transformer to predict the future trajectories of surrounding vehicle. Validation of the proposed algorithm is performed using the NGSIM dataset. The results demonstrate that the average intention prediction accuracy of the IDBN is improved by 4.22% over the Non-Interactive Dynamic Bayesian Network. Among four existing models, the IDBN exhibits the best performance. Compared to the Transformer without intention fusion, the root mean square error of the intention fusion Transformer decreases by 5.61%-10.8%. The study significantly advances the prediction and decision-making in automated driving, and therefore greatly promotes the transportation safety and automobile intelligence.