Stochastic Control Behavior of the Balancing Rider for Cycling Safety in Traffic Simulation

Cyclist models with realistic behavior are essential for simulation-based safety testing of automated vehicles, assistance systems, traffic control, and infrastructure. We investigated the role of bicycle dynamics and human control for traffic processes in an experiment with ten cyclists following unanticipated, visual heading commands under varying time-pressure at multiple speeds, and propose a new Balancing Rider model for traffic simulation. It combines the Carvallo-Whipple bicycle model with full-state feedback to capture bicycle dynamics, human heading control, and balancing. We estimated control parameters for each individual maneuver, and described their distribution in a stochastic model. We tested trajectories generated from this distribution for predictive power and simulated obstacle avoidance to analyze safety prediction in comparison to a planar baseline model. The cycling experiment revealed a large variety of slow and quick control strategies for heading changes. The Balancing Rider model achieved excellent calibration and promising predictive performance covering the observed control strategies. A baseline planar point model couldn't capture details of cyclist maneuvers like countersteering and heading overshoot, with path deviations reaching an order of meters. For obstacle avoidance, we project > 0.4 s Time-to-Collision difference between the models, showing that the added physical detail greatly impacts simulated safety assessment. This illustrates that bicycle dynamics and human control behaviors are important elements explaining causal cycling traffic processes and should be included in models for simulated safety testing. The large variety of control behaviors among the participants could potentially be an explanatory factor of traffic conflict outcomes, which future work should investigate further.