Agentic Simulation for Wildfire Evacuations

AgentEvac simulates a wildfire evacuation as a population of households deciding what to do. Each agent is driven by a large language model and reasons through the Protective Action Decision Model (PADM), the behavioural framework for how people notice a warning, interpret the threat it describes, and choose a protective action. Those decisions become vehicle trips inside SUMO, an open source traffic simulator, so the moment a household chooses to leave turns into load on a real road network.

Coupling the two puts behaviour and traffic in the same model instead of feeding a fixed departure curve into a network simulation. The aim is to study how warning timing, staged departures, and route guidance change the time a community needs to clear, and where a plan that looks sound on paper produces congestion that costs people the margin they were given.