Han, Dongming (Southeast University), Cheng, Sizhe (Southeast University), Wang, Jinxiang (Southeast University), Fang, Zhenwu (National University of Singapore), Chu, Duanfeng (Wuhan University of Technology), Yin, Guodong (Southeast University)

Social-Interaction-Aware Trajectory Prediction with Attention of Heterogeneous Traffic Participants

Scheduled for presentation during the Regular Session "S25a-Cooperative and Connected Autonomous Systems" (TH-LM-T25), Thursday, November 20, 2025, 12:10−12:30, Cooleangata 4

2025 IEEE 28th International Conference on Intelligent Transportation Systems (ITSC), November 18-21, 2025, Gold Coast, Australia

This information is tentative and subject to change. Compiled on October 18, 2025

Abstract

Trajectory prediction is a critical component for efficient decision-making and safe planning in autonomous driving systems. However, the behavioral uncertainty of heterogeneous participants in mixed-traffic environments poses significant challenges to driving safety. This paper proposes a social-interaction-aware trajectory prediction (SITP) method that incorporates the interaction features of heterogeneous traffic participants (HTPs). First, the temporal and spatial features are extracted in parallel using a Long Short-Term Memory (LSTM) network and a Graph Convolutional Network (GCN), respectively. By integrating social interaction information, including participants’ relative motion states and physical size, the extraction of spatial features can be enhanced. Second, an interaction attention module with interaction mask and absence mask is designed to extract the attention distribution toward the social interactions of different participants. Finally, a temporal decoder based on LSTM is employed to generate the predicted trajectory. The proposed method is trained and evaluated on the InD dataset. Experimental results demonstrate that the proposed SITP method achieves superior long-term prediction accuracy compared to baseline algorithms and exhibits strong adaptability across scenarios with varying levels of complexity.