Refining the Invisible: Cooperative Occlusion-Aware Planning for Connected and Automated Vehicles via Reachability Analysis

Achieving safe and efficient navigation in urban environments, where occlusions frequently occur, remains a persistent challenge for autonomous driving. Cooperative perception (CP) provides an effective way to enhance perception and mitigate occlusions. Nevertheless, for collision avoidance with potential hidden traffic participants (TPs) within residual occlusions shared among multiple connected agents, most studies still follow the worst-case assumption, forcing agents to remain overly conservative. In this paper, we propose a cooperative occlusion awareness model, considering numerous and spatially sparse occlusions in CP, as well as the impact of communication issues during information fusion for handling occlusions. It leverage shared perception information to perform occlusion tracking, aiming to refine the feasible state space of potential hidden TPs including both existence intervals and speed bounds, facilitating less conservative behavior while maintaining safety. The model first performs occlusion modeling to initialize potential hidden TP states within residual occlusions, then applies occlusion association to capture correspondences between occlusions across frames and surrounding TPs, and finally refines the state space by eliminating unreachable states based on the cooperatively perceived road context revealed through the association results. Simulations across diverse occlusion scenarios demonstrate the effectiveness of the proposed model over baselines relying solely on onboard sensors or object-level fusion, as well as its robustness to communication issues containing sensor position and pose errors, delays and interruptions.