Description
Agent-Based Models simulate crowd dynamics by giving each pedestrian an individual rule-set with internal state (goal, mood, awareness) and letting macroscopic patterns emerge from local interactions. ABMs subsume social-force-model, cellular-automata + floor-field-model, and rule-based heuristics, but the term is normally used when agents have richer cognition than pure physics-style particles. The thesis treats ABM as the umbrella category for microscopic crowd simulators.
When it's used
- Heterogeneous crowds (groups, families, mobility-impaired agents)
- What-if scenarios where individual decisions matter
- Coupling with
data-assimilationto fit real trajectories - BLE-derived trajectory replay for
fundamental-diagramcalibration
Limitations
- Computational cost grows quickly with agent count if each carries deep state
- Parameter inference is hard without trajectory ground truth
- Validation against real crowds requires careful metric choice