Occlusion-Aware Planning for Connected and Automated Vehicles with Cooperative Perception at Unsignalized Intersection

Achieving safe and efficient navigation in urban environments remains a significant challenge for autonomous driving, primarily due to frequent occlusions that hinder perception. Cooperative perception (CP) has emerged as a promising solution, offering enhanced sensing capabilities by enabling information sharing among vehicles. In this work, we propose an occlusion-aware motion planning framework that integrates CP to optimize vehicle speed, minimizing risk while ensuring efficient navigation. The proposed framework operates in a sequential pipeline. At each time step, sensor features, vehicle motion data, and contextual map information from an edge server are shared among vehicles to enable cooperative object tracking and occlusion analysis. To quantify the risks posed by occluded areas, we introduce a probabilistic representation of potentially hidden objects, which dynamically adapts to varying viewpoints. In addition, an association module is designed to establish correspondences between potentially hidden object and observed objects, thereby improving the accuracy of risk assessment. Finally, predictive information from both observation and motion spaces is incorporated into the planning and control module to guide reference speed planning and vehicle maneuvering. Extensive simulation results demonstrate that our approach significantly improves safety and driving efficiency in complex occlusion scenarios, outperforming baseline methods that rely solely on onboard sensors or single-view perception fusion.