文献类型：会议论文

英文题名：Occlusion-Aware Game-Theoretic Multi-Agent Planning for Connected Vehicles

作者：Li, Jiahong[1];Jiang, Beiyan[1]

第一作者：李佳洪

通讯作者：Li, JH[1]

机构：[1]Beijing Union Univ, Coll Robot, Beijing, Peoples R China

第一机构：北京联合大学机器人学院

通讯机构：[1]corresponding author), Beijing Union Univ, Coll Robot, Beijing, Peoples R China.|[1141739]北京联合大学机器人学院;[11417]北京联合大学;

会议论文集：9th CAA International Conference on Vehicular Control and Intelligence-CVCI

会议日期：OCT 24-26, 2025

会议地点：Qingdao, PEOPLES R CHINA

语种：英文

外文关键词：Connected and Automated Vehicles; Occlusion; Dynamic Games; Contingency Planning; Iterative LQ

摘要：Planning failures in connected and automated vehicles (CAVs) often occur when occlusions hide other traffic from the ego vehicle's sensors, and human drivers and cyclists pursue unknown objectives and react to the ego's behavior. Existing planners typically address these factors in isolation: visibility-aware methods turn over-cautious, whereas intent-aware contingency planners assume perfect visibility and miss hidden hazards. We propose Occlusion-Aware Contingency Games (OC-Games), a dynamic game-theoretic planner that handles both occlusions and intent uncertainty simultaneously. The timeline is segmented into alternating open-loop intervals (players cannot observe each other) and feedback intervals (full state information is shared). The ego plans a single trunk trajectory that branches automatically when its Bayesian belief about other agents' intents becomes confident enough. This unified game model captures temporary loss and recovery of line-of-sight, belief-driven strategy branching, and rational best responses by all agents. To deal with the intractable computation of finding exact equilibria, we develop a real-time iterative solver. An outer iterative LQ game loop repeatedly linearizes the nonlinear dynamics and costs. Each linear-quadratic subgame is then solved via a hybrid Riccati backward pass that seamlessly interleaves open-loop and feedback information structures. The solver uses belief entropy evolution to identify the branching moment and runs in cubic time in the joint state dimension. We demonstrate our planner in three challenging scenarios, i.e., rural overtaking, urban intersection, and highway weaving, and show that coupling visibility and intent reasoning yields behavior that is both assertive and safe, pointing toward a practical path for robust large-scale CAV deployment.