[Paper Review] Quantitative Verification of a Force-based Model for Pedestrian Dynamics

Mohcine Chraibi, Armin Seyfried+2 more|ArXiv.org|Dec 20, 2009

Evacuation and Crowd Dynamics4 references9 citations

TL;DR

This paper proposes a modified, spatially continuous force-based model for pedestrian dynamics that quantitatively reproduces empirical data on flow, density, and fundamental diagrams in corridors and bottlenecks. By simplifying the Centrifugal Force Model with a two-sided Hermite interpolation for repulsive forces and calibrating free parameters, the model achieves accurate simulation of pedestrian behavior without numerical instabilities or velocity restrictions, showing strong agreement with real-world measurements in both corridor and bottleneck scenarios.

ABSTRACT

This paper introduces a spatially continuous force-based model for simulating pedestrian dynamics. The main intention of this work is the quantitative description of pedestrian movement through bottlenecks and in corridors. Measurements of flow and density at bottlenecks will be presented and compared with empirical data. Furthermore the fundamental diagram for the movement in a corridor is reproduced. The results of the proposed model show a good agreement with empirical data.

Motivation & Objective

• To develop a force-based pedestrian dynamics model that enables quantitative validation against real-world data.
• To address numerical instabilities in existing models like the Centrifugal Force Model without requiring additional collision detection algorithms.
• To reproduce key empirical phenomena such as the fundamental diagram, flow through bottlenecks, and density distributions in confined spaces.
• To calibrate free parameters to achieve accurate, scenario-specific simulation results.

Proposed method

• The model uses Newton’s second law, with repulsive forces between pedestrians and obstacles, and a driving force toward a desired velocity direction.
• Repulsive forces are modeled as inversely proportional to distance, with a relative velocity term that prevents influence from faster pedestrians ahead.
• A two-sided Hermite interpolation is applied to the repulsive force to ensure smooth decay to zero at safe distances and prevent singularities at close range.
• The driving force is defined as a relaxation term toward a desired velocity with a time constant τ = 0.5 s.
• The system is solved numerically using an Euler scheme with a fixed time step Δt = 0.01 s.
• Free parameters (ν, da, db) are calibrated to match empirical data, and pedestrian mass is set to unity for simplicity.

Experimental results

Research questions

• RQ1 Can a spatially continuous force-based model quantitatively reproduce empirical fundamental diagrams in pedestrian corridors?
• RQ2 How accurately can the model simulate pedestrian flow and density at bottlenecks compared to real measurements?
• RQ3 Does the proposed modification eliminate numerical instabilities without requiring additional collision detection procedures?
• RQ4 To what extent does the model capture the difference in density between the bottleneck and the area in front of it?
• RQ5 Can the model be calibrated to reproduce empirical data across different corridor and bottleneck widths?

Key findings

• The model successfully reproduces the fundamental diagram in corridors of widths 1 m, 2 m, and 4 m, showing good agreement with empirical data from Mori1987, Helbing2007, Oeding1963, and Hankin1958.
• Flow measurements through bottlenecks of widths from 0.8 m to 1.2 m show strong quantitative agreement with empirical data from Seyfried2009b.
• Density in front of the bottleneck is significantly higher than inside the bottleneck, and the model correctly captures this difference and the amplitude of fluctuations.
• The simulated density profiles inside and in front of the bottleneck match empirical results from Seyfried2009b and Rupprecht2007.
• The model achieves accurate results without velocity restrictions or additional collision detection, thanks to the Hermite interpolation of the repulsive force.
• With calibrated parameters (ν=0.2, da=0.3 m, db=0.2 s), the model produces stable, realistic simulations across various scenarios.