We model the human factors
disaster simulations miss.
LLM agents simulate how people think · an agent-based model simulates how the world changes.
Agents evacuating on the real Noto road network.
Three assumptions existing models get wrong.
Real data: departure delays of 0–30 minutes, varying significantly by demographic group.
Under uncertainty, people follow crowds or head to familiar routes rather than optimal ones.
Rule-based ABMs cannot explain individual behaviour; planners cannot interpret model outputs.
A Disaster World Model driven by LLM agents.
Converts raw data into a structured world state, updated each timestep.
Reads the world state, reasons in natural language, and acts. Movement updates the state.
Matches simulated population patterns against real human-mobility data.
The 2024 Noto Peninsula event.
Three layers — Data → Environment → Agent — built on the real earthquake-tsunami: building damage, OSM roads, GSI/JSGS inundation, designated shelters, and the JMA warning timeline.
- 1Data Layer
Building damage · road network · tsunami inundation · shelters · warning timeline · run-up survey.
- 2Environment Engine
Converts the raw layers into a structured world state at each timestep — passability, flooding, capacity, warning level.
- 3LLM Agent Population
Survey-initialized personas Observe → Think → Act, emitting natural-language reasoning for every decision.
Run the whole loop, in one board.
Drive the pipeline and inspect what it discovered — the hidden variables, their causal effects, agent reasoning, and the calibration — live.