Deep Reinforcement Learning-based Context-Aware Redundancy Mitigation for Vehicular Collective Perception Services

Collective perception service (CPS) is one of the most fundamental services in intelligent transportation systems. Since it can incur significant overhead in exchanging perceived object containers (POCs), european telecommunications standards institute (ETSI) introduced several redundancy mitigation schemes; however, there are several limitations in application to the vehicular environment. In this paper, we propose a deep reinforcement learning (DRL)-based context-aware redundancy mitigation (DRL-CARM) scheme where various vehicular contexts (i.e., location, speed, heading, and perception area) are employed for redundancy mitigation. To derive the optimal policy on redundancy mitigation, the DRL-CARM scheme employs a deep Q-network (DQN) with a reward function on the usefulness of POC. Evaluation results demonstrate that the DRL-CARM scheme can improve the average usefulness of POC by 254% and reduce the network load by 49.4%, compared with conventional redundancy mitigation schemes.