Intelligent Task Offloading Decisions for Enhanced Swarm-Based UAV Surveillance

Unmanned aerial vehicles (UAVs) deployed in swarms are emerging as a promising solution for a wide range of surveillance tasks, including border monitoring, disaster response, and urban security. However, the effectiveness of UAV swarms is limited by the challenges of processing the high-computation demands of such surveillance tasks. Thus, to process high-computation and diverse surveillance tasks, offloading to nearby edge devices with powerful processing units or ground base stations (GBSs) offers reduced delay and energy consumption. However, determining the optimal computational offloading decision is challenging due to the diverse surveillance tasks and their individual delay and energy requirements. In this paper, we introduce a reinforcement learning approach, utilizing a trust region policy optimization with an actorcritic (TRAC) algorithm, to address the problem of the decision on whether the computation is offloaded to the edge computing server or GBSs. The proposed TRAC algorithm is designed to meet the stringent delay and energy requirements by ensuring efficient and timely processing of surveillance tasks. In addition, the edge server managing surveillance tasks from large numbers of swarm UAVs faces increased delays and energy consumption when processing each surveillance task till the end to get the outcome. Such a process fails to meet the delay and energy requirements due to significant waiting and processing. Thus, to further optimize delay and energy constraints, we implement a multi exit framework on edge servers, enabling early results from intermediate layers if a specified accuracy threshold is met, rather than processing tasks to completion. In addition, we employ Brakerski-Fan-Vaikuntanathan (BFV) homomorphic encryption to enhance the security of sensitive surveillance data. Simulation results demonstrate that the proposed TRAC-based offloading strategy, combined with the intermediate multi exit framework, outperforms state-of-the-art methods in reducing network delay by up to 61.2% and energy consumption by up to 57.2%.