Pedestrian Dynamics
JuPedSim Evacuation Analysis
A configurable pedestrian-evacuation workflow built around a real building geometry, multiple spawn regions, several exit-assignment strategies, and reproducible trajectory output.
Problem
Building-evacuation performance depends on where pedestrians start, which exits are available, how exits are assigned, and how quickly agents move. The project was developed to compare these factors in a controlled, repeatable simulation workflow.
My contribution
- Imported and prepared the walkable building geometry from a WKT file.
- Defined nine exit polygons and six configurable pedestrian spawn regions.
- Created six scenarios with different active spawns, exit assignments, and pedestrian speeds.
- Parallelised the initial agent-distribution step using
ThreadPoolExecutor. - Used Numba to accelerate nearest-exit selection for constrained exit groups.
- Exported trajectories to SQLite and generated scenario animations and runtime summaries.
Simulation methodology
Movement model
JuPedSim CollisionFreeSpeedModel with configurable desired speeds and motivation time.
Geometry
Real building walkable area imported from HC.wkt, with explicit exits and spawn regions.
Outputs
SQLite trajectories, evacuation times, simulation runtimes, configuration plots, and animations.
Scenario design
| Scenario group | Main comparison |
|---|---|
| Scenarios 1–2 | Same spawn region assigned to two different exits. |
| Scenarios 3–4 | Selected spawn combinations with logically constrained exit groups. |
| Scenarios 5–6 | All spawn regions active, comparing mean pedestrian speeds of 1.2 and 2.0. |
Representative visuals
Engineering outcome
The project provides a repeatable framework for comparing evacuation configurations instead of evaluating only one simulation. By separating geometry, spawn configuration, exit logic, movement parameters, and output generation, the workflow can be extended to additional scenarios and validation studies.
Limitations and next steps
Exit selection is based partly on Euclidean distance and scenario-defined logical constraints rather than a complete route-choice model. Future improvements include path-aware exit selection, stronger quantitative post-processing with PedPy, parameter calibration, and systematic sensitivity analysis.