LLM-Enabled Behavioral Simulation

LLM-Enabled Behavioral Simulation investigates how large language models can help specify, generate, and adapt human behavior within transparent simulation environments.

Overview

The project uses large language models as constrained components of a broader simulation workflow, not as unrestricted substitutes for behavioral models. LLMs help translate intervention goals into activity-level changes, generate plausible behavioral alternatives, and support rapid scenario design. Mechanistic rules, spatial constraints, and validation checks keep generated behavior grounded in the simulated environment.

Research Questions

• How can natural-language policy goals be translated into executable mobility interventions?
• Where should LLM reasoning end and mechanistic simulation begin?
• How can generated behaviors be constrained, audited, and compared across model versions?

Methods

• Prompted and structured behavior generation
• Agent-based simulation with rule-based safeguards
• Scenario comparison and intervention analysis
• Traceability, validation, and sensitivity testing

Applications

• Rapid design of mobility interventions
• Public-health and emergency-response scenarios
• Human-centered policy exploration
• Adaptive agents in urban digital twins

Selected Outputs

• Agile design of activity-based mobility interventions using LLMs
• Reusable workflows for grounded generative agents
• Evaluation protocols for LLM-assisted simulation