EnvSocial-Diff: A Diffusion-Based Crowd Simulation Model with Environmental Conditioning and Individual-Group Interaction

TL;DR

EnvSocial-Diff is a novel diffusion-based crowd simulation model that incorporates environmental factors and multi-level social interactions to produce more realistic pedestrian trajectories, outperforming current state-of-the-art methods.

Contribution

It introduces a structured environmental conditioning module and a graph-based individual-group interaction module, enhancing realism in crowd simulation.

Findings

Outperforms state-of-the-art crowd simulation methods

Effectively encodes environmental constraints and attractors

Captures interpersonal and group-level social dynamics

Abstract

Modeling realistic pedestrian trajectories requires accounting for both social interactions and environmental context, yet most existing approaches largely emphasize social dynamics. We propose \textbf{EnvSocial-Diff}: a diffusion-based crowd simulation model informed by social physics and augmented with environmental conditioning and individual--group interaction. Our structured environmental conditioning module explicitly encodes obstacles, objects of interest, and lighting levels, providing interpretable signals that capture scene constraints and attractors. In parallel, the individual--group interaction module goes beyond individual-level modeling by capturing both fine-grained interpersonal relations and group-level conformity through a graph-based design. Experiments on multiple benchmark datasets demonstrate that EnvSocial-Diff outperforms the latest state-of-the-art methods, underscoring the importance of explicit environmental conditioning and multi-level social interaction for realistic crowd simulation. Code is here: https://github.com/zqyq/EnvSocial-Diff.