Toward Optimal Resource Allocation: A Multi-Agent DRL Based Task Offloading Approach in Multi-UAV-Assisted MEC Networks

Abstract

The application of UAV-aided MEC well-suited for the execution of the data-intensive and latency-sensitive tasks in the infrastructure-deprived regions. However, the growing number of UAVs and smart devices causing a major difficulty in the devising an effective scheme for the task offloading and resource allocation in multi-UAV-aided MEC networks. Furthermore, the resource deficient environments unable to sustain prolonged resource-intensive activities, additional complexities are posed on the optimum utilization of the resources. In this paper, we introduced a multi-agent deep reinforcement learning scheme for the task offloading in the multi-UAV-assisted networks (MUAVDRL). In this configuration, the mobile users fetch computational resources from the UAVs with the goal of minimizing the computation cost which incorporates both the energy consumption and the computation delay. Initially, we start with the optimization problem which is defined as the minimizing the computational costs. Through modelling it as MDP, we aim to reduce the computational costs for mobile users. Leveraging the dynamic and high-dimensional nature of the challenge, the MUAVDRL algorithm solves this problem efficiently. Comprehensive simulation results exhibit the efficacy and superiority of our projected framework when compared to existing state-of-the-art methods, illustrating its potential in the practice.