Modeling experimental observations of cyclist conflict behavior in open spaces

Logo (Animatie) ©2025 Marbus, A., Konrad, C.M., Happee, R., Dabiri, A., Moore, J.K. Modeling experimental observations of cyclist conflict behavior in open spaces A. Marbus, C.M. Konrad, R. Happee, A. Dabiri, J.K. Moore Public version of the slideset presented by C. M. Konrad at ICSC 2025, 05.11.2025, Oslo, Norway, with minor adaptions for copyright. This work is licensed under a Creative Commons Attribution License (CC-BY). DOI: 10.5281/zenodo.17642627 ©2025 The Authors | https://doi.org/10.5281/zenodo.17642627 Alleen titel ©2025 The Authors | https://doi.org/10.5281/zenodo.17642627 Motivation Cyclist Interactions Unstructured Environments Conflict Causality Understand … For … Safer Infrastructure / Bicycles / Cars Cyclist Models for Simulated Testing Foto of cyclists on a busy open space. Removed in public version for data privacy. Examples of similar scenes: - https://www.groningenbereikbaar.nl/nieuws/samen-voorrang-bepalen-bij-alle-richtingen-groen - https://bicycledutch.wordpress.com/2016/04/19/utrecht-cycling-city-of-the-netherlands/ - https://dvhn.nl/groningen/Iedereen-tegelijk-groen-Wachten-duurt-lang-te-lang-24619115.html ©2025 The Authors | https://doi.org/10.5281/zenodo.17642627 Tekst ©2025 The Authors | https://doi.org/10.5281/zenodo.17642627 Background: Traffic Conflict Causality Initial Conditions U 𝑃(𝑢) Evasive Actions X 𝑃(𝑥|𝑢) Outcomes Y 𝑃(𝑦|𝑥, 𝑢) Davis et al., 2011 Position Orientation Speed Spatiotemporal Factors Roll Angle Countersteer Stabilization Bicycle Dynamics and Control Tekst ©2025 The Authors | https://doi.org/10.5281/zenodo.17642627 Method: Overview 4 Cycling Experiment Kinematic Data Analysis Adding Bicycle Dynamics Computational Interaction Model Alleen titel ©2025 The Authors | https://doi.org/10.5281/zenodo.17642627 Method: A Cycling Interaction Experiment 4 Interaction Angles 45° / 90° / 135° / 180° 10 Pairs of Cyclists Male - Female pairing 8 Interactions per Pair 16 Reference Runs per Cyclist without Partner Instrumented Bicycles (position, attitude, velocities) (Figure adapted from Marbus, 2025) (Figure from Marbus, 2025) Alleen titel ©2025 The Authors | https://doi.org/10.5281/zenodo.17642627 Method: A Cycling Interaction Experiment 4 Interaction Angles 45° / 90° / 135° / 180° 10 Pairs of Cyclists Male - Female pairing 8 Interactions per Pair 16 Reference Runs per Cyclist without Partner Instrumented Bicycles (position, attitude, velocities) Trial video (180° interaction) (Figure from Marbus, 2025) Tekst ©2025 The Authors | https://doi.org/10.5281/zenodo.17642627 Method: Overview 7 Cycling Experiment Kinematic Data Analysis Adding Bicycle Dynamics Computational Interaction Model Tekst 8 ©2025 The Authors | https://doi.org/10.5281/zenodo.17642627 Results: Kinematic Analysis - Paths Distributions differ between interaction and non-interaction Qualitatively different strategies per scenario Path adjustments (Figure adapted from Marbus, 2025) 45°90°135°180° NonInteraction Interaction Tekst 9 ©2025 The Authors | https://doi.org/10.5281/zenodo.17642627 Results: Kinematic Analysis ANOVA on path deviation: Interaction Non-interaction •F(1, 19) = 58.04, p < 0.001) significant •Pairwise 7/8 significant (except 135° right turn) ANOVA on average speed: Interaction Non-interaction •F(1, 19) = 5.76, p = 0.027) significant •Pairwise not significant Predominantly path adjustments instead of speed adjustments. path deviation [m] Paths speed [m/s] Speeds (Figures adapted from Marbus, 2025) Tekst ©2025 The Authors | https://doi.org/10.5281/zenodo.17642627 Outlook: Investigating Bicycle Dynamics and Rider Control Is this really all that is happening? Explicitly model bicycle dynamics and reactive rider control: → Probably not! Cyclists need to: •countersteer to roll the bike into the corner •stabilize the bicycle (from Stienstra et al., 2024, CC-BY) Preprint out soon! Tekst 17 ©2025 The Authors | https://doi.org/10.5281/zenodo.17642627 Outlook: Investigating Bicycle Dynamics and Rider Control Adding bicycle dynamics and rider control does not improve the predictions Perfect command follower Perfect follower median 𝐴𝐷𝐸𝐴 median 𝑀𝐴𝐸𝜓 Outliers (𝑀𝐴𝐸𝜓>15) train 0.65 m 7.89° 0.8 % test 0.63 m 7.53° 10.0 % Balancing rider model Balancing rider 0.73 m 9.75° 12.5 % Tekst 18 ©2025 The Authors | https://doi.org/10.5281/zenodo.17642627 Outlook: Investigating Bicycle Dynamics and Rider Control Perfect command follower Balancing rider model Measured Heading Predicted Heading Measured Heading Predicted Heading Bicycle Dynamics & Control Adding bicycle dynamics and rider control does not improve the predictions Perfect follower median 𝐴𝐷𝐸𝐴 median 𝑀𝐴𝐸𝜓 Outliers (𝑀𝐴𝐸𝜓>15) train 0.65 m 7.89° 0.8 % test 0.63 m 7.53° 10.0 % Balancing rider 0.73 m 9.75° 12.5 % Heading offset model describes emergent behavior, not underlying processes Tekst 19 ©2025 The Authors | https://doi.org/10.5281/zenodo.17642627 Outlook: Investigating Bicycle Dynamics and Rider Control Perfect command follower Balancing rider model Measured Heading Predicted Heading ??? Predicted Heading = Measured Heading Bicycle Dynamics & Control Adding bicycle dynamics and rider control does not improve the predictions Perfect follower median 𝐴𝐷𝐸𝐴 median 𝑀𝐴𝐸𝜓 Outliers (𝑀𝐴𝐸𝜓>15) train 0.65 m 7.89° 0.8 % test 0.63 m 7.53° 10.0 % Balancing rider 0.73 m 9.75° 12.5 % Heading offset model describes emergent behavior, not underlying processes Alleen titel ©2025 The Authors | https://doi.org/10.5281/zenodo.17642627 Key Take-Aways Christoph M. KonradPhD CandidateBiomechatronics Human-Machine ControlDelft University of [email protected] Christoph M. Konrad PhD Candidate Biomechatronics & Human-Machine Control Delft University of Technology [email protected] Reach Out! Open space cycling conflicts: ▪predominantly path adjustments over speed adjustments. ▪Crossing first ≠ taking evasive action ▪Cyclists evade to the outside if opponent in the direction of their turn (>90% of all cases). ▪Cyclists cut inwards if opponent not in direction of their turn (>90% of all cases). ▪Decision boundary depends on relative orientation. ▪Bicycle dynamics are necessary to understand underlying control behavior. Co-authors and supervisors Anna Marbus Riender Happee Jason Moore Azita Dabiri Experiment conducted as part of the Masters’s thesis: Marbus, A. (2025). Cyclist conflict behavior in shared spaces [Master Thesis, Delft University of Technology], https://repository.tudelft.nl/record/uuid:092f3b70-2d97-436e-b193-139a593e09c7 20 Alleen titel ©2025 The Authors | https://doi.org/10.5281/zenodo.17642627 References Davis, G. A., Hourdos, J., Xiong, H., & Chatterjee, I. (2011). Outline for a causal model of traffic conflicts and crashes. Accident Analysis & Prevention, 43(6), 1907–1919. https://doi.org/10.1016/j.aap.2011.05.001 Marbus, A. (2025). Cyclist conflict behavior in shared spaces. [Master Thesis, Delft University of Technology]. https://resolver.tudelft.nl/uuid:092f3b70-2d97-436e-b193-139a593e09c7 Stienstra, T., Brockie, S. & Moore, J. (2024). BRiM: A Modular Bicycle-Rider Modeling Framework [version 2; peer reviewed]. The Evolving Scholar - BMD 2023, 5th Edition. https://doi.org/10.59490/660179a06bf1082286458109 Reach Out! 21