Learning From Real-World Trajectories: A Hybrid Genetic Algorithm and Reinforcement Learning Approach for Vulnerable Road User Model Calibration [brochure]

Microscopic simulation models require accurate calibration to reflect real-world dynamics. Vulnerable road user (VRU) model calibration is challenging due to: Complex interactions and behaviors, Mode-specific interactions, Diversity in infrastructure and the existence of shared spaces. Existing calibration methods, relying solely on numeric loss functions, often fail to capture realistic, human-perceived behavior.