Reproduce “Emergent Crowds Dynamics from Language-Driven Multi-Agent Interactions” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Yibo Liu, Liam Shatzel, Brandon Haworth, Teseo Schneider

University of Victoria researchers developed a language-driven multi-agent framework that integrates Large Language Models (LLMs) to control individual agent movement, bridging social interaction with physical locomotion. This method allows agents to make context-aware decisions based on dialogue, vision, and internal states, leading to emergent, unscripted crowd behaviors such as grouping, information propagation, and dynamic redistribution.

Key insights

  • LLMs can effectively bridge the gap between abstract language-based social simulation and concrete physical motion control in multi-agent crowd systems, enabling context-aware decision-making.
  • Conditioning individual agent behavior on personality, emotional state, and social relationships, alongside local perceptual inputs, facilitates the emergence of richer and more natural crowd dynamics.
  • Decentralized, language-driven control over an agent's entire locomotion pipeline (goal, path, steering) allows for unscripted and generalizable complex social interactions and macro-level crowd behaviors.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.