Intelligent orchestration of computation and networking for drone swarm applications
Abstract
[EMBARGOED UNTIL 5/1/2024] Drone swarms that are equipped with high-resolution sensors and video cameras can benefit many applications (e.g., disaster response management, smart farming, traffic control). This requires sensor localization and access to edge computation and networking resources to provide environmental situational awareness in the applications. Orchestration of the computation and networking resources in practice are performed in isolation, and do not sufficiently account for factors such as limited battery life of drones that impacts geographical area coverage in application missions. Moreover, prior works focus mainly on computation offloading, and do not account for the factors that impact drone swarm communication such as intermittent network link failures and environmental obstacles (e.g., physical buildings, strong winds) that impact video transmission, data transfers and thus disrupt the application mission plans. To address above knowledge gaps, this innovative research dissertation aims to investigate a trans-formative co-design of localizing, computation offloading, and control networking in drone swarms that serve diverse application missions. The research goal is to first develop awareness models in terms of application environments with sensors and drone mobility, which can then guide the creation of edge computation and networking resource orchestration algorithms in an intelligent and joint manner. To this end, specific novel research objectives include: (i) develop awareness models for system awareness (energy consumption), environment awareness (physical obstacle, wind), and mobility awareness (ground sensor location, drone trajectory) in drone swarm application scenarios; (ii) create intelligent computation offloading and control networking algorithms to leverage the awareness models to ensure optimal scheduling of data processing tasks across edge, cloud and network resources to meet application mission requirements; and (iii) implement intelligent algorithms in a Drone Swarm Localizing, Computation Offloading, Control and Networking framework viz., "DroneCOCoNet" to validate the integration benefits in drone-aided applications such as disaster response management and smart farming. Evaluation shows that our intelligent orchestration strategies build with the system, environment, and mobility models benefits: i) computation offloading procedure which achievement at least 240 percent higher median episode reward than baseline non-intelligent scheme, and ii) the adaptive on the network and video properties with at least 91 percent of throughput performance of the oracle baseline approach with at least 32 dB value (good video quality as perceived by users) for PSNR after transmission. This dissertation also includes the development of drone-network testbeds in terms of trace-based, learning-based, and scientific workflow supported. In addition to this, the application use cases on how drone swarms can benefit in various scenarios, i.e., disaster management, last-mile parcel delivery, and smart farming are also addressed.
Degree
Ph. D.