Jin, Yuhui (State Grid Shandong Electric Power Research Institute), Liu, Donglan (State Grid Shandong Electric Power Research Institute), siyang, li (Hangzhou International Innovation Institute of Beihang Universit), Gao, Rui (Beihang University), Liu, Xin (State Grid Shandong Electric Power Research Institute), Li, Chenxi (Beihang University), Huang, Jian (Beihang University)

HHIM: Heterogeneous and Hierarchical Interaction Modeling for Multiscale Intention and Trajectory Prediction

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

Trajectory prediction plays a vital role in improving the safety of autonomous driving vehicles and understanding of human driving behaviour. The prediction task is full of challenges mainly because the movement of the target is affected by all kinds of surrounding objects. Therefore, how to correctly model the social interactions and scene constraints is the key factor. In addition, more than one plausible and socially acceptable future paths exists. This high degree of uncertainty must also be considered. To address the heterogeneity of multiagents in dynamic social systems, we first propose a hierarchical heterogeneous graph representation and interaction modeling method (HHIM) to model heterogeneous relationships between different types of agents and different semantic relationships between the same type of agents. Vector-level and agent-level interactions are modelled separately in the type-specific polyline subgraph and the directed heterogeneous graph. Category nodes in the directed heterogeneous graph help to explicitly model the consistency of motion patterns within the same category. As for multimodal problems, we propose a multiscale intention space to serve as a different level of abstraction of future intention estimation and to alleviate the uncertainty. Extensive experimental results on public datasets demonstrate that our method outperforms the state-of-the-art baseline methods, validating the feasibility and effectiveness of the developed approach.