Multi-Agent Based Crowd Evacuation Simulation
Jan 2026
Finished
Multi-agent crowd evacuation simulation built on Mesa (Python agent-based modeling framework). Defines 3 agent types with distinct behaviors: polite (yield, crowd-sensitive), aggressive (push, low deadlock sensitivity), and slow (reduced speed, high density sensitivity). The environment uses a discrete grid (OrthogonalMooreGrid) with 6 scenarios: OPEN, MALL, CORRIDOR, SEATS (theater seats), SNAKE (winding path), and RANDOM. Agents use configurable pathfinding (Manhattan, BFS, A*) to navigate toward exits. Detailed metrics are collected: evacuation rate, macro/micro speed, local density, deadlock factor — both per agent type and globally. Massive batch experiments (1000 runs per config, 16 parallel processes) analyze 7 dimensions: agent density, aggressive impact, slow agent impact, pathfinding algorithms, exit preference, number of exits, and exit width. Includes interactive Jupyter widget visualization and a LaTeX paper.
AI
Jupyter
LaTeX
Matplot
NumPy
Pandas
Plotly
Python
