TinyDeepUAV: A Tiny Deep Reinforcement Learning Framework for UAV Task Offloading in Edge-Based Consumer Electronics

Abstract

Recently, there has been a rise in the use of Unmanned Areal Vehicles (UAVs) in consumer electronics, particularly for the critical situations. Internet of Things (IoT) technology and the accessibility of inexpensive edge computing devices present novel prospects for enhanced functionality in various domains through the utilization of IoT-based UAVs. One major difficulty of this perspective is the challenges of computation offloading between resource-constrained edge devices, and UAVs. This paper proposes an innovative framework to solve the computation offloading problem using a multi-objective Deep reinforcement learning (DRL) technique. The proposed approach helps in finding a balance between delays and energy consumption by using the concept of Tiny Machine Learning (TinyML). It develops a low complexity frameworks that make it feasible for offloading tasks to edge devices. Catering to the dynamic nature of edge-based UAV networks, TinyDeepUAV suggests a vector reinforcement that can change weights dynamically based on various user preferences. It is further conjectured that the structure can be enhanced by Double Dueling Deep Q Network (D3QN) for optimal improvement of the optimization problem. The simulation results depicts a trade-off between delay and energy consumption, enabling more effective offloading decisions while outperforming benchmark approaches.