ITSC 2025 Paper Abstract

Paper VP-VP.25

Wang, Anzheng (Wuhan University of Technology, Wuhan 430063, China), Chu, Duanfeng (Wuhan University of Technology), Deng, Zejian (University of Waterloo), Lu, Liping (Wuhan University of Technology), Wang, Jinxiang (Southeast University), Sun, Chen (Univiersity of Hong Kong)

Psychological Field-Encoded Trajectory Prediction with a Future Path-Guided Decoder for Autonomous Driving at Signalized Intersections

Scheduled for presentation during the Video Session "On-Demand Video Presentations" (VP-VP), Saturday, November 22, 2025, 08:00−18:00, On-Demand Platform

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 April 2, 2026

Keywords Real-time Motion Planning and Control for Autonomous Vehicles in ITS Networks, AI, Machine Learning Techniques for Traffic Demand Forecasting, AI, Machine Learning for Dynamic Traffic Signal Control and Optimization

Abstract

Abstract-Trajectory prediction is a critical task in autonomous driving, particularly at signalized intersections where dynamic traffic light controls and diverse vehicle behaviors pose significant challenges to existing methods. In this paper, we propose an enhanced Graph Neural Network (GNN)-based trajectory prediction approach that uses Psychological Field-Encoding and a Future Path-Guided Decoder (PFPD). In the encoding stage, the psychological field models the influence of traffic lights on vehicle behavior by mapping traffic light phase information into control degree scores. We further employ map self-attention to capture the road network's structure by encoding lane adjacency relationships. A Graph Attention Network (GAT) is utilized to perform temporal encoding, agent-lane encoding, and agent-agent encoding, capturing complex interactions. For decoding, our Future Path-Guided Decoder applies cross-attention over temporal, lane, and spatial features to capture contextual dependencies. It is followed by a path-to-mode attention mechanism that aligns embedded future path features with mode prototypes, effectively guiding trajectory decoding. Experimental results on the V2X-Seq and SinD datasets demonstrate that PFPD achieves state-of-the-art performance across all evaluation metrics, including Average Displacement Error (minADE6), Final Displacement Error (minFDE6), and Miss Rate (MR).