Joint Optimization of UAV Trajectory and Task Scheduling in SAGIN: Delay Driven

Abstract

This paper aims to study a computation task scheduling problem in the space-air-ground integrated network (SAGIN). The prior works on this problem usually assume that an unmanned aerial vehicle (UAV) is static or has a fixed flying trajectory. In this paper, we allow a UAV to plan its own trajectory and to have a certain coverage area. Our objective is to design a policy that minimizes the maximum task processing delay by joint optimization of task scheduling and UAV trajectory. We first formulate this nonconvex optimization problem as a Constrained Markov Decision Process (CMDP) under the constraints of UAV energy capacity and mobility space. Then, we design a Deep Deterministic Policy Gradient (DDPG)-based reinforcement learning algorithm to learn the optimal task offloading ratio and UAV trajectory. Our work is evaluated from three aspects: (1) SAGIN network architecture vs. single layer network; (2) DDPG-based algorithm vs. Deep Q Network-based algorithm; (3) optimized UAV trajectory vs. fixed UAV position. Experiment results validate that the optimized UAV trajectory can achieve a lower task processing delay than fixed UAV position .