Abstract
Accurate trajectory prediction in multi-agent environments is fundamental to understanding and forecasting collective motion in complex scenes. However, existing methods often struggle to capture the dynamic nature of interactions and higher-order group relationships. To address these challenges, we propose a novel framework that seamlessly integrates pairwise and group-wise interaction modeling. Our approach introduces a Step-wise Pairwise Interaction Modeling (SPIM) module to effectively capture dynamic interaction features over time, and a Hypergraph-based Global Interaction Modeling (HGIM) module to effectively model higher-order group relationships using adaptive hypergraph structures. Through the integration of these complementary modules within a unified multimodal decoder architecture, our method achieves competitive performance on the SDD and ETH-UCY datasets while significantly reducing model complexity. These results highlight the superior efficiency and effectiveness of our proposed approach.
Funding Information
  • General Program of National Natural Science Foundation of China (52172380)
  • Key Research and Development Project of Shaanxi Province (2024GX-ZDCYL-01-25)
  • Key Research and Development Project of Shaanxi Province (2024CY-JJQ-55)

This publication has 31 references indexed in Scilit: