Task Offloading and Resource Allocation in UAV-Aided Emergency Response Operations via Soft Actor Critic

Abstract

In modern emergency responses, unmanned aerial vehicles (UAVs) play a crucial role in redefining disaster management through diverse task execution (e.g., object detection). However, UAVs are usually resource-constrained. To address this issue, UAV mother-ships with edge servers (UMECs) can be placed near UAV scouts for remote task processing. UMECs can have a larger scale of energy consumption and battery capacity than UAV scouts. Therefore, minimizing the total or maximum energy consumption may result in ignoring the energy consumption in UAV scouts and focusing on UMECs, consequently reducing the network’s lifetime. Furthermore, UMECs might lack the capacity to process all types of tasks owing to memory or software restrictions, and faster task execution often necessitates both central processing unit (CPU) and graphics processing unit (GPU), which are rarely considered by existing works. Therefore, this paper studies a task offloading and resource (computation capacity and power) allocation (TORA) problem with the goal of minimizing the maximum energy consumption ratio among UMECs and UAV scouts alongside task execution latency ratio and total energy consumption, where tasks are executed using both CPU and GPU, each UMEC can execute a specific set of tasks, and devices have limited resources. We then formulate the problem as a non-convex mixed-integer nonlinear programming problem and decompose it into multiple sub-problems. Finally, a soft-actor-critic (SAC) based TORA algorithm (SATORA) is proposed to address the problem, which can adapt to the time-varying environment scenario. Numerical simulation results show that SATORA outperforms other baseline algorithms.

Keywords

• Deep reinforcement learning
• resource allocation
• task offloading
• multi-access edge computing