Service Placement and Trajectory Design for Heterogeneous Tasks in Multi-UAV Cooperative Computing Networks

TL;DR

This work addresses energy-efficient computation offloading in a multi-UAV MEC network with heterogeneous task types. It reformulates the problem as a constrained Markov Decision Process and introduces SAC-TORA, a centralized Soft Actor-Critic framework that jointly optimizes task scheduling, service placement, UAV relaying, resource allocation, and 3D trajectories to minimize the long-term energy $E_{tot}$. The approach demonstrates superior energy savings compared with baselines such as DDPG and PPO, and outperforms fixed-service and equal-allocation schemes, highlighting the practical benefits of flexible learning-based coordination in UAV-assisted MEC. The contribution offers a scalable, robust method for reducing energy consumption in dynamic MEC environments, with potential impact on energy-aware autonomous networks and disaster-response deployments.

Abstract

In this paper, we consider deploying multiple Unmanned Aerial Vehicles (UAVs) to enhance the computation service of Mobile Edge Computing (MEC) through collaborative computation among UAVs. In particular, the tasks of different types and service requirements in MEC network are offloaded from one UAV to another. To pursue the goal of low-carbon edge computing, we study the problem of minimizing system energy consumption by jointly optimizing computation resource allocation, task scheduling, service placement, and UAV trajectories. Considering the inherent unpredictability associated with task generation and the dynamic nature of wireless fading channels, addressing this problem presents a significant challenge. To overcome this issue, we reformulate the complicated non-convex problem as a Markov decision process and propose a soft actor-critic-based trajectory optimization and resource allocation algorithm to implement a flexible learning strategy. Numerical results illustrate that within a multi-UAV-enabled MEC network, the proposed algorithm effectively reduces the system energy consumption in heterogeneous tasks and services scenarios compared to other baseline solutions.

Figures (10)

• Figure 1: The proposed multi-UAV cooperative MEC network model.
• Figure 2: Diagram of SAC-TORA algorithm implementation.
• Figure 3: The convergence of SAC-TORA and two baseline algorithms.
• Figure 4: Comparison of performance across different numbers of users.
• Figure 5: Comparison of performance across various numbers of UAVs.