Li, Jinze (Nanyang Technological University), Liu, Haochen (Nanyang Technological University), Zhao, Bolin (Nanyang Technological University), Yang, Haohan (Nanyang Technological University), Lou, Baichuan (Nanyang Technological University), Lv, Chen (Nanyang Technological University)

Structured Motion Prediction with Efficient Spatial-Temporal Graph Attention for Autonomous Racing

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

Abstract

Accurate and physically-consistent motion pre-diction is crucial for autonomous racing, where race cars must anticipate the future states of competitors under high-speed, dynamic conditions with minimal margin for error. This paper introduces an efficient learning-based prediction framework for autonomous racing that jointly models agent behavior and racetrack geometry. Central to our approach is a real-time spatiotemporal interactive predictor design. A dual-stream recurrent encoder captures temporal motion dynamics and extract spatial context from track boundaries via graph convolutions. An efficient structured motion decoder that predict physically consistent and geometrically aligned trajectories with the racetrack layout. The proposed framework is evaluated against several strong racing predictors on the benchmarked real-world MixNet dataset. Results demonstrate superior performance, with high prediction accuracy while consistently producing smooth and track-compliant trajectories. The design prioritizes both interpretability and physical feasibility, enabling real-time deployment in racing scenarios.