Melo Castillo, Angie Nataly (University of Alcala), Salinas Maldonado, Carlota (University of Alcala), Sotelo, Miguel A. (University of Alcala)

Anticipating the Invisible: A Knowledge-Based Agent for Occluded Pedestrian Collision Avoidance in Virtual Scenarios

Scheduled for presentation during the Regular Session "S27a-Safety and Risk Assessment for Autonomous Driving Systems" (TH-LM-T27), Thursday, November 20, 2025, 11:50−12:10, Broadbeach 3

2025 IEEE 28th International Conference on Intelligent Transportation Systems (ITSC), November 18-21, 2025, Gold Coast, Australia

This information is tentative and subject to change. Compiled on October 18, 2025

Abstract

Poor pedestrian visibility is recognized as one of the common risk factors contributing to pedestrian crashes. Despite advances in perception algorithms and collision avoidance systems, there remains a significant gap in handling scenarios involving fully occluded pedestrians. In this work, we present a custom behavior agent that leverages a knowledge-based occluded pedestrian predictor to influence the behavior of an Autonomous vehicles (AV). We designed and implemented 72 scenarios in Virtual Reality (VR) to evaluate the performance of both the agent and the predictor, using metrics such as collision avoidance rate, Time To Collision (TTC), and Pedestrian Detection Anticipation Time (PDAT). We compared the results against traditional agents that rely on standard detection algorithms. Preliminary findings demonstrate the effectiveness of incorporating contextual information to predict fully occluded pedestrians: Our agent achieved an average pedestrian detection anticipation time of 4.27 seconds and a collision avoidance rate of 87.5% across all experiments.