Fast and Fair Computation Offloading Management in a Swarm of Drones Using a Rating-Based Federated Learning Approach

• Published in: IEEE access Vol. 9; pp. 113832 - 113849
• Main Authors: Rahbari, Dadmehr, Alam, Muhammad Mahtab, Moullec, Yannick Le, Jenihhin, Maksim
• Format: Journal Article
• Language: English
• Published: Piscataway IEEE 2021; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Collaboration; collaborative computing; Computation offloading; Computational modeling; Computer architecture; Drone aircraft; Drone vehicles; Drones; Edge computing; Energy consumption; Federated learning; Mobile computing; multi-access edge computing; Network latency; Object recognition; Real time; Resource utilization; Servers; Swarm of drones; Task analysis; Unmanned aerial vehicles; Video data
• ISSN: 2169-3536; 2169-3536
• DOI: 10.1109/ACCESS.2021.3104117

Today, unmanned aerial vehicles (UAVs) or drones are increasingly used to enable and support multi-access edge computing (MEC). However, transferring data between nodes in such dynamic networks implies considerable latency and energy consumption, which are significant issues for practical real-time applications. In this paper, we consider an autonomous swarm of heterogeneous drones. This is a general architecture that can be used for applications that need in-field computation, e.g. real-time object detection in video streams. Collaborative computing in a swarm of drones has the potential to improve resource utilization in a real-time application i.e., each drone can execute computations locally or offload them to other drones. In such an approach, drones need to compete for using each other's resources; therefore, efficient orchestration of the communication and offloading at the swarm level is essential. The main problem investigated in this work is computation offloading between drones in a swarm. To tackle this problem, we propose a novel federated learning (FL)-based fast and fair offloading strategy with a rating method. Our simulation results demonstrate the effectiveness of the proposed strategy over other existing methods and architectures with average improvements of −23% in energy consumption, −15% in latency, +18% in throughput, and +9% in fairness.