Task-Aware Multi-UAV Cooperative Computing: A Learning-Based Trajectory and Resource Optimization

Abstract

Aerial computing is an important form of mobile edge computing (MEC) to enhance network coverage. In this paper, we focus on a massive access scenario where ground devices with different types of tasks are not in service area of communication infrastructure. Therefore, we propose a task-aware multiple unmanned aerial vehicles (UAV) cooperative computing scheme that each UAV stores the program for executing a certain type of tasks. To minimize the completion time of all tasks, we formulate a problem that jointly considers trajectory design and computation resource allocation for multiple UAVs, as well as user access decision, while guaranteeing quality of services. As the problem is a mixed-integer non-convex optimization which is difficult to solve, we propose a multi-agent deep reinforcement learning-based approach, where the multi-agent deep deterministic policy gradient (MADDPG) algorithm is applied. Considering the high-dimensional continuous action space, a particle swarm optimization (PSO) algorithm for access decision and resource allocation is introduced to reduce the complexity. Simulation results show that the proposed multi-UAV cooperative computing method has a better effect than baseline approaches on reducing the total completion time.