Wei, Manman (University of Tsukuba), Onishi, Masaki (National Institute of Advanced Industrial Science and Technology), Yin, Yingjie (Toyota Technical Development Corp.)

Prediction of Multi-Pedestrian Interaction with Vehicles in Shared Spaces: A Particle-Based Graph Neural Network Approach

Scheduled for presentation during the Regular Session "S02b-Optimization for Shared, Electric, and Sustainable Mobility Systems" (WE-EA-T2), Wednesday, November 19, 2025, 14:10−14:30, Southport 2

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 19, 2025

Abstract

Modeling pedestrians' motion is essential for the safe navigation of autonomous vehicles in shared spaces. Pedestrian movement is inherently complex, characterized by intricate interactions and frequent transitions between behavioral modes, making accurate prediction challenging. Rather than predicting a single trajectory, modeling the potential spatial distribution of pedestrian locations is more meaningful, as autonomous vehicles are primarily concerned with the likely positions of pedestrians along a specific path. Although existing methods based on deep generative models have improved prediction accuracy by incorporating uncertainty, they often encounter challenges such as limited model interpretability and unreasonable predictions. To address these issues, this paper proposes a novel approach that integrates graph neural networks (GNNs) with a particle-based modeling framework. By leveraging the particle-based method, the proposed model captures various possibilities of pedestrian behavior, aiming to enhance both the rationality and interpretability of trajectory predictions. The effectiveness of the proposed method is validated using publicly available datasets, demonstrating its capability to produce more reliable and informative trajectory prediction distributions. The proposed model enables safer and more reliable autonomous navigation in shared spaces by accurately predicting diverse and interaction-aware pedestrian trajectories.