Abstract
Task offloading technology plays a crucial role in the Internet of Vehicles (IoV) by minimizing processing delays through the joint optimization of heterogeneous computing resources supported by vehicles, roadside units (RSUs), and macro base stations (MBSs). Previous works have often ignored the wireless interference during the exchange and sharing of task data. Additionally, the potential for vehicles with similar driving behaviors to form vehicle platooning (VEH-PLA) and effectively integrate individual vehicle resources has not been adequately addressed. Furthermore, as a novel resource management paradigm, VEH-PLA should consider task categorization since vehicles within a VEH-PLA may have identical task offloading requestsan aspect that has also received insufficient attention. In this paper, considering wireless interference, vehicle mobility, VEH-PLA, and task categorization, we propose four task offloading models aimed at minimizing processing delays. By utilizing centralized training and decentralized execution (CTDE) based on multi-agent deep reinforcement learning (MADRL), we present a task offloading decision-making method to find the global optimal offloading decision. This results in significant enhancements in resource load balancing and reductions in processing delays. Finally, simulations validate that the proposed method significantly outperforms traditional task offloading approaches in terms of minimizing processing delays while maintaining balanced resource utilization.
Funding Information
  • National Natural Science Foundation of China (62301076, 62341101, 62321001)
  • Fundamental Research Funds for the Central Universities (24820232023YQTD01)
  • Macao Young Scholars Program (AM2023015)
  • Natural Science Foundation of Beijing Municipality (L232003)
  • National Key Research and Development Program of China (2022YFB4300403)
  • Science and Technology Development Fund, Macau SAR (0188/2023/RIA3)

This publication has 40 references indexed in Scilit: