Joint Trajectory and Resource Optimization for Multi-UAV Cooperative Computation

• Published in: IEEE International Conference on Communications workshops pp. 1611 - 1616
• Main Authors: Xu, Wenlong, Zhang, Tiankui, Mu, Xidong, Liu, Yuanwei, Wang, Yapeng, Shi, TianYi
• Format: Conference Proceeding
• Language: English
• Published: IEEE 09.06.2024
• Subjects: Approximation algorithms; Autonomous aerial vehicles; Benchmark testing; Conferences; deep reinforcement learning; Heuristic algorithms; mobile edge computing; multi-UAV cooperative computation; resource allocation; Simulation; Trajectory; trajectory planning
• ISSN: 2694-2941
• DOI: 10.1109/ICCWorkshops59551.2024.10615538

In the multiple unmanned aerial vehicle (UAV) mobile edge computing (MEC) systems, the cooperative computation among multiple UAVs can improve the overall computation service capability. Multi-UAV MEC systems can meet the quality of service requirements for computation intensive applications of ground terminals (GTs) in complex field environments, emergency disaster relief and other special scenarios. In this paper, a multi-UAV cooperative computation framework is proposed while taking the GT movement and random arrival of computation tasks into consideration. A long-term optimization problem is formulated for the joint optimization of UAV trajectory and resource allocation, subject to minimizing the total GT computation task completion time and the total system energy consumption. To solve this problem, a joint multiple time-scale optimization algorithm is proposed. In particular, the optimization problem is decomposed into a long time-scale multi-UAV trajectory planning subproblem and a short time-scale resource allocation subproblem. The proximal policy optimization algorithm is invoked to solve the long time-scale subproblem. The greedy algorithm and the successive convex approximation (SCA) method are employed to solve the short time-scale subproblem. Simulation results show that the proposed algorithm can quickly adapt to different degrees of environmental dynamics and outperforms the benchmark algorithm for different task requirements and available resources.