🚀 Data-Driven Framework for Agent-Based Pedestrian Modeling and Simulation
Overview:
This project presents a data-driven computational framework for calibrating and validating heterogeneous agent-based pedestrian models using real-world multi-agent trajectory data.
By integrating Particle Swarm Optimization (PSO) with a deep learning surrogate model (TabNet), the framework accelerates parameter calibration while maintaining accuracy in reproducing pedestrian behaviors in various environments.
Figure: Snapshots of pedestrian simulation at different steps.
🔍 Output & Results
The framework enables:
Efficient calibration of multiple agent parameters from real-world data.
Simulation outputs that closely match actual pedestrian velocities and trajectories.
Scalability for testing different environment layouts and crowd densities.
Validation results showed strong alignment between simulated and observed pedestrian flow patterns, reducing calibration runtime significantly compared to traditional brute-force methods.
📄 Details & Resources
For full methodology, and dataset details, please visit:
S. Amartaivan and H. Morita,
"Efficient Calibration of Heterogeneous Agent-Based Pedestrian Model Using PSO With Deep Learning Surrogates",
IEEE Access, vol. 13, pp. 109185-109199, 2025, doi: 10.1109/ACCESS.2025.3582778.
