AI-Enabled Collaborative Distributed Computing in Networked UAVs

Abstract

Nowadays, the evolution of AI is noticed in supporting many life applications and manipulating different data types. It helps complete the tasks and get the required information efficiently and precisely. The deployment of AI techniques and machine learning models moves to limited-resources energy-constrained platforms, ranging from simple IoT devices to unmanned aerial vehicles (UAVs). Employing such models on limited-resource UAVs to support a wide range of applications is an inevitable duty and it is at the same time a challenging task. Additionally, obtaining high-accuracy outputs from a single AI-enabled UAV within the operating context of delay-sensitive applications faces a lot of obstacles, and may not be feasible. Accordingly, distributed operations and cooperation among a set of UAVs can provide the required level of accuracy within the time constraints for some applications. This work proposes a distributed computing architecture for networked UAVs based on collaborative learning and edge-of-things computing. Such architecture would help a suite of UAVs to train based on their local ML model and captured data and to collaborate with other UAVs in the same network to generate an aggregated ML model that improves the operation accuracy with acceptable performance speed. Using a networked UAV system and various application scenarios, numerical simulation studies have been presented. The performance analysis and results show how the proposed distributed computing architecture with collaborative learning outperforms the centralized computing architecture with edge and cloud computing paradigms.

Keywords

• Artificial intelligence
• collaborative learning
• lightweight training
• networked UAVs
• resource allocation
• vehicular edge of things computing