Enabling High-Performance Onboard Computing with Virtualization for Unmanned Aircraft Systems
Abstract: Unmanned Aircraft Systems (UAS) have gained increasing popularity in both academia and industry, as reflected by its wide range of applications in search and rescue, package delivery, fire detection and so on. In these applications, computation-intensive tasks such as image processing are often fulfilled by offloading the tasks from the UAS of limited processing capabilities to the powerful ground stations or cloud servers. However, the vulnerable air-to-ground communication link of limited bandwidth can cause significant transmission delays or even failures. To address this issue, delay-sensitive tasks should be performed directly onboard of UAS. In this study, we explore suitable microcomputers and virtualization techniques to enable high-performance onboard computing for UAS. In particular, the capabilities of various microcomputers were first analyzed and compared from the aspects critical to UAS onboard computing, which provides guidelines for microcomputer selection. Two representative virtualization schemes, the Kernel-based Virtual Machine (KVM) and Docker, were then implemented on a selected microcomputer Jetson TX2, which equip UAS with powerful resource management capabilities. To understand the impact of the two virtualization techniques on the performance of UAS, extensive experiments were conducted to evaluate their performances from various aspects that are critical to UAS operations, including computing, networking, and isolation, etc. The results show the promising performance of Docker over KVM in computing, networking, and isolation of most hardware resources, as well as the weakness of Docker in security. Moreover, to understand the benefits of virtualization in facilitating real UAS applications, various UAS applications including image processing, object tracking and distributed computing were also investigated. To the best of our knowledge, this is the first work in the literature that systematically investigates the microcomputer selection and virtualization for UAS.
Baoqian Wang* and Junfei Xie
Judging Forms – Official judges only