Article Details

Abstract

Driving intention recognition is a key element of Intelligent Transportation Systems (ITS) and autonomous driving decision modules. Although traditional deep learning methods have good performance in data-driven prediction, they lack physical constraints, which easily leads to unreasonable prediction scenarios. Therefore, this paper proposes a method introducing Physics-Informed Neural Networks (PINN) into trajectory prediction, and constructs a PL-LSTM framework fusing Long Short-Term Memory (LSTM). By introducing a physical loss function containing vehicle dynamics, driving behavior priors, and traffic rules, it imposes constraints on the prediction process. Experiments based on the NGSIM dataset confirm that PL-LSTM significantly outperforms baseline methods in terms of accuracy and trajectory error (ADE/FDE). Ablation experiments further analyze the importance of different physical constraints. PL-LSTM can achieve smoother and more reasonable trajectory predictions that conform to traffic regulations. The relevant research conducted in this paper provides new means for intention recognition in multi-agent driving scenarios.

Keywords

driving intention recognition; NGSIM; multi-agent learning; PINN; LSTM; autonomous driving

References

1. [1] Long Haoming, Xue Zhenfeng, Chen Zhuo, etc Overview of trajectory prediction methods for intelligent driving vehicles [J]. Ordnance Automation, 2024, 43 (5): 43-52
2. [2] Fang Huazhen, Liu Li, Gu Qing, etc Research Status of Lane Changing Intent Recognition in Autonomous Vehicles [J]. Journal of Engineering Science, 2024, 46 (10): 1845-1855
3. [3] Yu Li, Li Huiyuan, Jiao Chenlu, etc Pedestrian trajectory prediction in complex scenes based on multi head attention adversarial mechanism [J]. Journal of Computer Science, 2022, 45 (6): 1133-1146
4. [4] Ji Xuewu, Fei Cong, He Xiangkun, etc Driving intention recognition and vehicle trajectory prediction based on LSTM network [J]. Chinese Journal of Highways, 2019, 32 (6): 34-42
5. [5] WANG J Q，LIU K，LI H T，et al. Vehicle Trajectory Prediction Using Hierarchical LSTM and Graph Attention Network [J].IEEE Internet of Things Journal，2025，12（6）：7010-7025.
6. [6] DEO N，TRIVEDI M M. Multi-Modal Trajectory Prediction of Surrounding Vehicles with Maneuver Based LSTMs [C]//2018 IEEE Intelligent Vehicles Symposium(IV).Changshu：IEEE，2018：1179-1184.
7. [7] Zhao Shuen, Su Tianbin, Zhao Dongyu Interactive Vehicle Driving Intent Recognition and Trajectory Prediction Based on Graph Neural Networks [J]. Automotive Technology, 2023 (7): 24-30
8. [8] RHINEHART N，MCALLISTER R，LEVINE S. Deep imitative models for flexible inference， planning， and control [C]//Proceedings of International Conference on Learning Representations(ICLR).Addis Ababa：ICLR，2020：1-12.
9. [9] DI X，SHI R Y，MO Z B，et al. Physics-Informed Deep Learning for Traffic State Estimation: A Survey and the Outlook [J/OL].Algorithms，2023，16（6）：305[2025-06-28].https://doi.org/10.3390/a16060305.
10. [10] Wang Mengqian, Cai Yingfeng, Wang Hai, etc Vehicle trajectory prediction method based on graph convolutional interactive network [J]. Automotive Engineering, 2024, 46 (10): 1863-1872
11. [11] Chu Duanfeng, Liu Hongxiang, Gao Bolin, etc A review of research on vehicle queue prediction cruise control [J]. Journal of Mechanical Engineering, 2024, 60 (18): 218-246

About the authors

Mei Yuankun (2002-), male, Han ethnicity, from Suizhou, Hubei, master's student, research interests include deep learning.