Graph-Based Pedestrian Locomotion Model

MCML Authors

Ana Cukarska

→ Group Felix Dietrich
 Physics-enhanced Machine Learning

Felix Dietrich

Prof. Dr.

Collaborating PI

Physics-enhanced Machine Learning

Abstract

We propose a data-driven approach to pedestrian locomotion modeling by combining ideas<br>from traditional knowledge-based models and machine learning. Our approach considers a human crowd<br>as a locally connected graph of neighbors. The data-driven algorithm incorporates this neighborhood<br>information to predict the next step of individual pedestrians. Unlike most machine learning approaches, our model is based on random features and thus can be trained rapidly, making real-world applications and rapid prototyping viable.

inproceedings CD26

---

TGF 2026

16th International Conference on Traffic and Granular Flow. Bristol, UK, Jun 16-19, 2026. To be published. Preprint available.

---

Authors

A. Cukarska • F. Dietrich

---

Links

PDF

---

Research Area

 A2 | Mathematical Foundations