PlaceSim: An LLM-based Interactive Platform for Human Behavior Simulation in Physical Facilities

Lee, S., Y. Yu, D. Shin, & R. Singh

Proceedings of the ACM International Conference on Information and Knowledge Management (CIKM), 2025

Seoul, Republic of Korea · doi:10.1145/3746252.3761461

Published

Abstract

Physical facility design faces a cold-start problem: predicting how people will behave in spaces that do not yet exist. Surveys capture stated preferences that diverge from actual usage, and existing simulation tools demand technical expertise that limits who can use them. PlaceSim is a web-based platform that uses large language models to simulate facility behavior through a zero-code interface, structuring model reasoning with a Persona-Environment-Scenario (P.E.S.) framework and context-aware AI personas whose decisions remain inspectable.

The platform supports interactive facility design, persona generation, live simulation with reasoning visualization, and what-if comparison across scenarios. Evaluated on 18 months of real apartment facility data covering 789,238 usage records from 8,435 residents, the zero-shot approach reaches Jensen-Shannon Divergence scores as low as 0.006, outperforming both supervised learning methods and existing LLM-based tools without requiring any training data.

Research Motivation

🎯 Problem Statement

Traditional human behavior simulation in physical spaces relies on simplified agent-based models that fail to capture the complexity of real human decision-making processes. Current approaches struggle with:

Limited contextual understanding of spatial environments
Inability to model complex social interactions and preferences
Lack of adaptability to diverse facility types and user demographics

🚀 Research Opportunity

Large Language Models present an unprecedented opportunity to create more realistic and nuanced simulations by:

Leveraging natural language understanding for complex behavioral reasoning
Incorporating rich contextual information about spaces and user intentions
Enabling dynamic adaptation to different scenarios and environments

Methodology

🏗️ Platform Architecture

PlaceSim integrates multiple components to create a comprehensive simulation environment:

LLM-based Reasoning Engine: Core decision-making component for agent behavior
Spatial Analytics Module: Processing and understanding of physical facility layouts
Behavioral Modeling Framework: Integration of psychological and sociological factors
Interactive Visualization: Real-time simulation display and control interface

🔬 Technical Implementation

Multi-Agent System: Each simulated person as an independent LLM-powered agent
Prompt Engineering: Specialized prompts for spatial reasoning and behavior prediction
Environmental Context Integration: Real-time facility information processing
Validation Framework: Comparison with real-world behavioral data

Key Findings

📊

Enhanced Realism

LLM-based agents demonstrate significantly more realistic behavior patterns compared to traditional rule-based simulations.

🎯

Contextual Adaptability

The platform successfully adapts to different facility types (offices, malls, museums) with minimal parameter adjustments.

⚡

Interactive Insights

Real-time simulation enables immediate feedback for facility design decisions and urban planning scenarios.

🔄

Scalable Framework

The modular architecture allows for easy extension to new environments and behavioral factors.

Applications & Impact

🏢 Urban Planning

Optimize building layouts and public space design based on predicted human flow patterns.

🏪 Retail Design

Improve store layouts and customer experience through behavior simulation.

🚨 Emergency Planning

Test evacuation procedures and safety protocols in virtual environments.

📚 Research Tool

Enable researchers to study human-space interactions at scale.

Technical Specifications

Large Language Models Simulation Modeling Spatial Analytics Multi-Agent Systems Interactive Visualization Behavioral Psychology Urban Planning