Dynamic Game-Theoretic Risk Assessment for Vehicle-Pedestrian Interaction With Perception Uncertainty

Pedestrians, as vulnerable road users, critically influence traffic safety and efficiency through their interactions with autonomous vehicles. Perception biases in judging vehicle distance and approaching velocity often lead to unsafe behaviors (e.g., risky crossing). However, existing studies based on game theory did not consider pedestrians' cognitive bias and sensor bias. To address this gap, we propose a dynamic risk assessment framework that integrates perceptual uncertainties. The model quantifies the effects of pedestrian visual biases and sensor noise, establishing an incomplete information game with Bayesian belief predication. The proposed model considers pedestrians' limited reasoning capacity characterized by two-stage discrete decision, while vehicles perform real-time observation and adjustment through continuous belief prediction. Dynamic conflict risk is evaluated by statistically analyzing the probability of mutual non-yielding between both agents at the second decision stage. Simulation experiments validate the model's efficacy, with results demonstrating its superior capability in reflecting real-world conflict risk.

Keywords: dynamic risk assessment, incomplete information game, bayesian belief predication, perception uncertainties.