MCORA-V2X: Conditionally-Enhanced MAPPO for Computation Offloading and Resource Allocation in V2X Networks

Abstract

Mobile Edge Computing (MEC) is recognized as an essential enabler in Intelligent Transportation Systems (ITS), providing communication and computation resources for vehicular applications. Advances in vehicular technology and the rise of autonomous vehicles, with their high mobility, dense traffic, and ultra-low-latency requirements, have increased the demand for communication and computational resources. Vehicle applications generate communication and computation-intensive tasks that are offloaded to adjacent vehicles or Roadside Units (RSUs) for processing. Task offloading and resource allocation remain significant issues due to high mobility, ultra-low latency, and high requirements for computation and communication resources. To address these shortcomings, this study presents MCORA-V2X, a conditionally-enhanced Multi-Agent Proximal Policy Optimization (MAPPO) framework for computation offloading and resource allocation in V2X networks. Unlike Vanilla-MAPPO formulations that employ a flat action space, MCORA-V2X introduces a conditional hierarchical actor that selects the execution tier: local, V2V, RSU/MEC, or cloud via a categorical head, and then allocates CPU and bandwidth through destination-specific policy heads conditioned on that choice. This hierarchical structure enables destination-aware resource allocation and provides a structured representation of heterogeneous resource requirements across offloading destinations. This V2X offloading problem is modeled as a multi-agent system and addressed using MCORA-V2X under the Centralized Training and Decentralized Execution (CTDE) paradigm. Comprehensive simulations were performed, and the results show that MCORA-V2X outperforms existing offloading schemes in raw episode reward, running average reward, average task delay, throughput, task drop rate, and task success rate.