RADELN
Bicycle infrastructure & network design – a human-centric, data-driven approach using spatio-temporal machine learning
Principle Investigators
Prof. Dr. Klaus Bogenberger,
Chair of Traffic Engineering and Control, Department of Mobility Systems Engineering
Prof. Dr. Stephan Günnemann,
Chair of Data Analytics and Machine Learning, Department of Informatics
Project contributors
Dr.-Ing. Simone Weikl,
Chair of Traffic Engineering and Control, Department of Mobility Systems Engineering
Victoria Dahmen, M.Sc.,
Chair of Traffic Engineering and Control, Department of Mobility Systems Engineering
Dominik Fuchsgruber, M.Sc.,
Chair of Data Analytics and Machine Learning, Department of Informatics
Motivation and Goals
RADELN aims to develop a data-driven framework to optimize bicycle infrastructure in cities, encouraging a modal shift from car to bike travel. By using GPS-based trip data we work on understanding the impact of infrastructure changes on bicycle use and route choice. By benchmarking against established methods and by using this knowledge to identify network bottlenecks, we will develop a framework that offers a phased plan for improving bike networks in Munich.
Recent Results
In the past year, we developed a bicycle routing model that provides insight into cyclists’ preferences and aids in estimating the flow per link in the network. We studied an analytical algorithm based on weighted shortest paths (WSP), a geometric Graph Neural Network (GNN), and a hybrid approach that combines both. While the hybrid approach most accurately recovers true trajectories only the probabilistic GNN captures the stochasticity of real paths and can make multiple predictions for a source-destination pair. As a next step, we will study the sensitivity of each model to infrastructure interventions, enabling informed decision making.
Selected documentation
Published/accepted journal/conference papers | Fuchsgruber, Dominik*; Dahmen, Victoria*; Bogenberger, Klaus; Günnemann, Stephan. Route Prediction for Mobility Data using Geometric Deep Learning. GNI Symposium & Expo on Artificial Intelligence for the Built World 2024. Dahmen, Victoria; Weikl, Simone; Bogenberger, Klaus: Interpretable Machine Learning for Mode Choice Modeling on Tracking-Based Revealed Preference Data. Transportation Research Record. 2024. |
Submitted journal / conference papers | Dahmen, Victoria; Loder, Allister; Bogenberger, Klaus: A novel Machine-Learning, Multi-Criteria, Centralized, Bicycle Routing Algorithm. Transportation Research Record. Fuchsgruber, Dominik; Postuvan, Tim; Günnemann, Stephan; Geisler, Simon. Graph Neural Networks for Edge Signals: Orientation Equivariance and Invariance. International Conference on Learning Representations 2025. |