Dynamic-Group-Aware Networks for Multi-Agent Trajectory Prediction with Relational Reasoning

TL;DR

This paper introduces DynGroupNet, a dynamic-group-aware network that models time-varying, group, and pairwise interactions among agents for improved multi-agent trajectory prediction, outperforming existing methods across multiple datasets.

Contribution

The paper proposes DynGroupNet, a novel model capable of dynamic relational reasoning and group interaction modeling without supervision, enhancing trajectory prediction accuracy.

Findings

DynGroupNet captures time-varying group behaviors and interaction categories.

It outperforms state-of-the-art methods with significant improvements in ADE/FDE metrics.

The model achieves state-of-the-art results on ETH-UCY dataset.

Abstract

Demystifying the interactions among multiple agents from their past trajectories is fundamental to precise and interpretable trajectory prediction. However, previous works mainly consider static, pair-wise interactions with limited relational reasoning. To promote more comprehensive interaction modeling and relational reasoning, we propose DynGroupNet, a dynamic-group-aware network, which can i) model time-varying interactions in highly dynamic scenes; ii) capture both pair-wise and group-wise interactions; and iii) reason both interaction strength and category without direct supervision. Based on DynGroupNet, we further design a prediction system to forecast socially plausible trajectories with dynamic relational reasoning. The proposed prediction system leverages the Gaussian mixture model, multiple sampling and prediction refinement to promote prediction diversity, training stability and trajectory smoothness, respectively. Extensive experiments show that: 1)DynGroupNet can capture time-varying group behaviors, infer time-varying interaction category and interaction strength during trajectory prediction without any relation supervision on physical simulation datasets; 2)DynGroupNet outperforms the state-of-the-art trajectory prediction methods by a significant improvement of 22.6%/28.0%, 26.9%/34.9%, 5.1%/13.0% in ADE/FDE on the NBA, NFL Football and SDD datasets and achieve the state-of-the-art performance on the ETH-UCY dataset.