5G边缘计算在无人机导航中的应用与卸载策略

需积分: 9 170 浏览量更新于2024-08-26 1 收藏 809KB PDF 举报

"这篇文章探讨了在5G网络中边缘计算对于无人机导航的作用，特别是针对使用摄像头导航的小型无人机。文章通过一个基于视觉的导航算法，分析了将图像处理任务从无人机上卸载到边缘服务器的策略。研究比较了三种计算模式：在无人机本地处理（onboard）、完全卸载到边缘服务器（fully offloaded）和部分卸载（partially offloaded）。部分卸载模式被认为对通信网络的传输速率要求较低，但仍需部分在无人机上进行处理。该研究的结果有助于根据网络条件选择最合适的图像处理模式，为决策是否卸载以及卸载哪些任务提供了依据。文章已被接受发表在未来的期刊上，目前是预印本版本。" 在5G时代，边缘计算正逐渐成为解决计算密集型任务的关键技术，特别是在无人机导航这样的实时性和低延迟要求高的应用中。无人机的计算能力通常有限，这限制了它们使用摄像头进行高速和灵活导航的能力。本文重点关注的是如何利用边缘计算来缓解这个问题。边缘计算是一种分布式计算架构，它将数据处理任务推送到网络的边缘，靠近数据生成的地方，而不是全部上传到云端处理。在无人机导航场景中，这意味着将图像处理任务从无人机移到更强大的边缘服务器上，从而减轻无人机的计算负担。文章提出了三种计算模式： 1. **板载(onboard)**：所有的图像处理都在无人机本地完成。这种方式不需要依赖网络，但在处理复杂任务时可能会受限于无人机的计算能力。 2. **完全卸载(fully offloaded)**：所有图像处理任务都转移至边缘服务器。这可以释放无人机的计算资源，但可能增加对网络带宽的需求，因为所有数据都需要实时传输。 3. **部分卸载(partially offloaded)**：只将部分图像处理任务卸载到边缘服务器。这种方法试图平衡计算能力和网络传输之间的需求，减少了对网络带宽的压力，同时保留了部分本地处理能力，以应对网络延迟或中断。通过实验和分析，作者发现部分卸载模式在通信效率和计算性能之间找到了一个较好的平衡点。这种模式可以根据网络状况和任务需求灵活调整，对于5G网络环境下的无人机导航可能是最具潜力的解决方案。边缘计算与5G的结合为无人机自主导航带来了新的机遇。5G网络的高速率和低延迟特性使得实时数据处理成为可能，而边缘计算则可以提供足够的计算资源，以支持复杂的视觉算法。通过优化卸载策略，可以提高无人机的导航性能，实现更快速、更准确的飞行控制。这篇文章的研究结果为5G环境下的无人机导航策略提供了理论依据，为未来无人机设计和网络架构优化提供了参考。随着边缘计算和5G技术的不断发展，我们可以期待更智能、自主的无人机系统将在物流、监控、搜索救援等多个领域发挥更大的作用。

2377-3766 (c) 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/LRA.2021.3062319, IEEE Robotics

and Automation Letters

IEEE ROBOTICS AND AUTOMATION LETTERS. PREPRINT VERSION. ACCEPTED FEBRUARY, 2021 1

Edge Computing in 5G for Drone Navigation: What

to Ofﬂoad?

Samira Hayat

, Roland Jung

, Hermann Hellwagner

1,2

, Christian Bettstetter

2,3

Driton Emini

, and Dominik Schnieders

Abstract—Small drones that navigate using cameras may be

limited in their speed and agility by low onboard computing

power. We evaluate the role of edge computing in 5G for such

autonomous navigation. The ofﬂoading of image processing tasks

to an edge server is studied with a vision-based navigation algo-

rithm. Three computation modes are compared: onboard, fully

ofﬂoaded to the edge, and partially ofﬂoaded. Partial ofﬂoading is

expected to pose lower demands on the communication network

in terms of transfer rate than full ofﬂoading but requires some

onboard processing. Our results on the computation time help

select the most suitable mode for image processing, i.e., whether

and what to ofﬂoad, based on the network conditions.

Keywords—Aerial Systems: Perception and Autonomy, Au-

tonomous Vehicle Navigation, Vision-Based Navigation.

I. INTRODUCTION

UTONOMY is desirable in many robot systems. In terms

of navigation, it dictates onboard perception and compu-

tations. The concept of self-driving cars has spurred research

in autonomous navigation. Improvements in sensor accuracy,

perception algorithms, high onboard processing power, and

low-latency communication have been topics of interest. The

existing solutions remain infeasible for unmanned aerial vehi-

cles (UAVs), or drones, as explained in the following.

Navigation autonomy in drone systems is essential in ap-

plications employing multiple drones such as disaster re-

sponse [1]. Small multicopter drones weighing less than a

kilogram are desirable candidate platforms due to their high

maneuverability; both moments of inertia and angular ac-

celeration scale with the characteristic dimension [2]. The

maneuverability introduces opportunities in exploration and

mapping of obstacle-ridden environments (e.g., ﬂight through

narrow vertical gaps [3]); the small size and payload capacity

limits the onboard compute power. High-speed vision-based

Manuscript received: October, 15, 2020; Revised January, 11, 2021; Ac-

cepted February, 8, 2021.

This paper was recommended for publication by Editor Pauline Pounds

upon evaluation of the Associate Editor and Reviewers’ comments. This work

was partially supported by Magenta Telekom (T-Mobile Austria GmbH) and

Deutsche Telekom AG, Germany, as well as by University of Klagenfurt (Karl

Popper Kolleg NAV), Austria.)

Samira Hayat (samira.hayat@aau.at) and Hermann Hellwagner are

with the Institute of Information Technology (ITEC), University of Klagenfurt.

Hermann Hellwagner, Roland Jung, and Christian Bettstetter are members

of the Karl Popper Kolleg on Networked Autonomous Aerial Vehicles (NAV),

University of Klagenfurt.

Christian Bettstetter is with the Institute of Net-

worked and Embedded Systems (NES), University of Klagenfurt.

Driton

Emini and Dominik Schnieders are with Deutsche Telekom AG.

Digital Object Identiﬁer (DOI): see top of this page.

navigation techniques targeting autonomous driving fail to

meet the challenges posed by small drone systems: high

acceleration and 3D mobility coupled with low capabilities

in terms of payload and onboard computation [4]. The six-

degree-of-freedom trajectories comprised of high linear and

angular velocities for high-speed navigation, such as in drone

racing, are not yet solved by the existing state estimation

solutions. The state-of-the-art methods deployed in practice

use simulations to learn control and perception policies for

navigation [5], [6]. Such learning-based solutions require

extensive simulations. In most cases, they can only tackle

static scenarios. For accurate trajectory generation in a speciﬁc

scene with moving obstacles, training on multiple static scenes

(with different obstacle placements) is needed to tackle the

dynamics. Higher speed compromises the navigation accuracy

due to the limited onboard computation capabilities [6].

This letter explores the role of edge computing to facilitate

high-speed vision-based autonomous navigation. We study

the ofﬂoading of the computationally intensive estimation of

poses based on sensor measurements to the edge server. Such

ofﬂoading requires a communication link with low latency

and high throughput to achieve real-time operation. New

possibilities arise with standardization activities in the Third

Generation Partnership Project (3GPP), tackling the integration

of drones into cellular networks [7]. The promised latency

and throughput ﬁgures are sufﬁcient to counter the mentioned

challenges. A drone may respond to its environment in near

real-time by sending the onboard generated sensor information

to an edge server and receiving corresponding state estimation

and control commands.

We evaluate the beneﬁts of using edge servers in enabling

real-time vision-based autonomous navigation of drones. To

this end, we use experimental results on 5G drone con-

nectivity [8] and proﬁling results of a standard monocular

Visual-Inertial Odometry (VIO) algorithm. Three modes are

compared: no ofﬂoading, partial ofﬂoading, and full ofﬂoading

of the image processing tasks to the edge server. With full

ofﬂoading, the edge server executes the entire image process-

ing pipeline, which requires the transfer of the full images

to the server, resulting in low computation burden onboard

and high communication demands. With partial ofﬂoading,

image features are detected and tracked onboard; the drone

then transfers the features to the edge server, thus ofﬂoading

the remaining image processing. This yields more onboard

computations than full ofﬂoading and lower communication

demands. All computations are performed onboard the drone

in the no ofﬂoading mode. The three modes are compared with

Authorized licensed use limited to: Stanford University. Downloaded on March 12,2021 at 01:02:08 UTC from IEEE Xplore. Restrictions apply.

下载后可阅读完整内容，剩余7页未读，立即下载

w_zhanglu

粉丝: 3
资源: 9

5G边缘计算在无人机导航中的应用与卸载策略

"计算机组成与结构：lecture 17 并行处理.pdf

"TI-SN75DP120.pdf: DP中继器解决方案详解

"研究云计算下的偏微分方程并行计算问题：架构、程序设计和实验环境

Collaborative Mobile Edge Computing in 5G Networks

5G网络安全技术探究.pdf

5G和4G网络优化.pdf

5G系统标准发展概述.pdf

《5G消息技术白皮书》.pdf

中国电信：面向5G的网络创新PPT.pdf

MEC边缘云加速5G行业数字化转型.pdf

最新资源