Multi-Agent Proximal Policy Optimization Based Redundancy Mitigation Rule for C-V2X Collective Perception

Abstract

Cooperative Perception (CP) enhances the perception capability of Connected and Automated Vehicles (CAVs) by sharing sensor information via Collective Perception Messages (CPMs). However, redundant transmissions of identical object information from multiple vehicles can lead to communication overload and inefficient resource usage. To address this issue, we propose a Multi-Agent Proximal Policy Optimization (MAPPO)-based Redundancy Mitigation Rule (RMR) that dynamically selects which objects to transmit based on each agent's local observation and shared policy. The proposed method is trained under a Centralized Training with Decentralized Execution (CTDE) framework using a shared actor and centralized critic. Simulation results demonstrate that our approach provides superior environmental awareness compared to existing ETSI RMR methods.