Distributed Queue-Aware Cooperative Scheduling Strategy for Multi-UAV Networks

Abstract

This paper investigates joint computation resource allocation and cooperative scheduling in distributed multi-UAV networks with parallel dual-task processing. Each UAV must simultaneously handle UAV-native sensing tasks and user-offloaded computation tasks, which compete for limited onboard computing resources and produce coupled, stochastic queue dynamics. Because partial observability and lack of global state information make centralized optimization impractical due to high communication overhead and limited scalability, myopic slot-wise decisions also fail to optimize long-term system performance. Therefore, we formulate a queue-aware, long-term optimization problem that jointly considers computation resource partitioning and inter-UAV task forwarding to reduce queue congestion and improve load balancing. To solve this, we propose a dynamic, distributed, consensus-assisted multi-agent proximal policy optimization (DDC-MAPPO) framework. In this framework, neighboring UAVs exchange lightweight local information to estimate network-wide load trends and support adaptive distributed coordination, enabling effective collaborative scheduling with limited information. Simulation results demonstrate that, compared with baseline and conventional MAPPO schemes, DDC-MAPPO achieves faster convergence, higher stable rewards, lower average and maximum queue backlogs, and better load balancing. Furthermore, as network scale increases, its performance gain becomes more significant, demonstrating strong effectiveness and scalability for distributed multi-UAV systems.