An LLM-powered Agent Framework for Heterogeneous Evacuation Behavior Modeling under a Moving Threat in a Public Plaza
Abstract
Modeling heterogeneous evacuation behavior under a moving threat is difficult because human perception, memory, and evidence evaluation are not well captured by fixed rules. We propose a novel LLM-powered agent-based framework to represent these internal decision processes. Each pedestrian agent perceives a private symbolic ASCII view, maintains a Memory-based Knowledge Graph derived solely from individual observations, and makes decisions through persona-conditioned prompts under a common sampling configuration. A compressed decision context with stateless memory preserves trial-and-error experience across turns while excluding reasoning traces, and a validation engine separates behavioral choice from physical feasibility by executing routes only over observed terrain. We evaluated eight personality compositions in eight paired randomized blocks within a simulated public plaza. Usable-exit knowledge was strongly associated with evacuation success: 89.5% of agents possessing such knowledge evacuated, compared with 1.05% of those without it. Personality compositions also differed in their evaluation of remembered threat evidence: the proportion of danger assessments varied by 0.265, while high-urgency, low-directness decisions ranged from 11.41% to 34.33%. After direct threat sightings, responses converged, with 99.6% of assessments classifying the situation as dangerous. Movement was selected in 99.8% of decisions. Overall, evacuation outcomes were strongly associated with information access, while evacuation time was jointly associated with spatial geometry, information, and affect. The framework provides an auditable approach to generating endogenous behavioral heterogeneity through persona-conditioned LLM agents in crowd-evacuation simulations.
Related
- A Dual Vibrational Probe Framework for Studying Solvation in Highly Heterogeneous Molecular Environments
- A Differentiable Framework for Inverse Identification of Heterogeneous Cohesive Laws
- Multiscale modeling of thermo-hydromechanical behavior of clayey rocks and application to geological disposal of radioactive waste
- Q-RAIL: A Reliability-Aware Framework for Quantum Federated Learning on Heterogeneous Noisy Hardware
- An Agent-Based Digital Twin for Energy Management in Ski Resorts: Integrating Skier Behavior Modeling with Lift Energy Consumption Optimization
BibTeX
@misc{ma2026llmpowered,
title = {{An LLM-powered Agent Framework for Heterogeneous Evacuation Behavior Modeling under a Moving Threat in a Public Plaza}},
author = {Jian Ma and Runxin Yu and Tianyu Tang and Xiaolian Li},
howpublished = {arXiv},
year = {2026},
month = sep,
doi = {10.48550/arxiv.2609.37009},
eprint = {2609.37009},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2609.37009},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - An LLM-powered Agent Framework for Heterogeneous Evacuation Behavior Modeling under a Moving Threat in a Public Plaza
AU - Jian Ma
AU - Runxin Yu
AU - Tianyu Tang
AU - Xiaolian Li
PY - 2026
DA - 2026/09/29
PB - arXiv
DO - 10.48550/arxiv.2609.37009
UR - https://arxiv.org/abs/2609.37009
AB - Modeling heterogeneous evacuation behavior under a moving threat is difficult because human perception, memory, and evidence evaluation are not well captured by fixed rules. We propose a novel LLM-powered agent-based framework to represent these internal decision processes. Each pedestrian agent perceives a private symbolic ASCII view, maintains a Memory-based Knowledge Graph derived solely from individual observations, and makes decisions through persona-conditioned prompts under a common sampling configuration. A compressed decision context with stateless memory preserves trial-and-error experience across turns while excluding reasoning traces, and a validation engine separates behavioral choice from physical feasibility by executing routes only over observed terrain. We evaluated eight personality compositions in eight paired randomized blocks within a simulated public plaza. Usable-exit knowledge was strongly associated with evacuation success: 89.5% of agents possessing such knowledge evacuated, compared with 1.05% of those without it. Personality compositions also differed in their evaluation of remembered threat evidence: the proportion of danger assessments varied by 0.265, while high-urgency, low-directness decisions ranged from 11.41% to 34.33%. After direct threat sightings, responses converged, with 99.6% of assessments classifying the situation as dangerous. Movement was selected in 99.8% of decisions. Overall, evacuation outcomes were strongly associated with information access, while evacuation time was jointly associated with spatial geometry, information, and affect. The framework provides an auditable approach to generating endogenous behavioral heterogeneity through persona-conditioned LLM agents in crowd-evacuation simulations.
ER -