Description
Understanding and predicting how people egress from a building or venue under emergency or near-emergency conditions: evacuation time, bottleneck formation, exit choice, panic regimes. This is the highest-stakes special case of crowd-dynamics, and a primary driver of building-code regulation. Evacuation modeling consumes BLE/WiFi-derived occupancy and flow estimates as boundary conditions in design-time simulations.
Why it's hard
- True emergency data is ethically off-limits; models calibrate against drills that under-represent panic regimes.
- Faster-is-slower and arch-and-clog non-linearities mean small parameter changes flip evacuation outcomes.
- Multi-storey, multi-exit topologies explode the state space; route-choice strategies dominate the result.
- Information dissemination (alarms, signage, smartphone alerts) is part of the dynamics yet rarely modeled jointly.
- Coupling with infectious-disease or smoke propagation is increasingly required.
Common approaches
- Cellular automata and lattice gas models for fast multi-storey simulation.
- Social Force-driven agent simulations for high-fidelity dynamics.
- Spatial-kinetic continuum models with disease/contagion coupling.
- Reinforcement learning for adaptive route-choice agents.
- Dijkstra / floor-field potential approaches for shortest-path routing under load.
Source Papers
- agnelli2023_ea3a ↗ — spatial kinetic model with infectious-disease contagion.
- sun2021_1423 ↗ — density-field + social-force evacuation simulation.
- lu2026_fb07 ↗ — extended cellular automaton for multi-storey evacuation.
- maury2018_d24a ↗ — Crowds in Equations (continuum framing).
- alattas2020_0e4e ↗ — LADM-IndoorGML for evacuation exercise tracking.