边缘智能驱动的高能效无人机自主导航算法研究

doi:10.11959/j.issn.2096-3750.2021.00229

摘要/Abstract

摘要：

近年来，无人机的自主导航技术在多个行业中受到了广泛的关注，相比于传统的导航技术，采用图像感知的深度学习方法具有很好的泛化能力并且不受全球定位系统（GPS, global positioning system）信号的影响，被证明是一种具有前景的自主导航方法。然而，深度学习的推断需要较大功耗，这对于能耗资源十分有限的无人机来说是一项挑战。针对该问题，基于边缘智能理论，将强化学习技术引入无人机端侧的推断过程中，根据无人机所处的环境复杂度实时感知信息，动态配置卷积神经网络的结构参数，使得无人机在保持稳定导航的同时，尽可能地减少计算功耗开销，实现无人机高可靠、低时延与高能效的自主导航飞行能力。该算法在仿真环境和现实环境中分别进行了验证，实验结果表明，相比于对比算法，所提的基于强化学习动态配置算法能够让无人机花费更少的计算能耗开销具有更长的飞行距离与更高的成功率。

关键词: 无人机, 边缘智能, 深度学习, 强化学习, 自主导航

Abstract:

Autonomous drone navigation has received growing attention in the recent community.Compared with traditional navigation approaches which rely on location-based services highly, deep learning based visual methods have showed superior performance in self-adaption and generalization, which are a promising solution for autonomous navigation.Running the resource-hungry deep learning execution in the resource-constrained unmanned aerial vehicle (UAV), however, significant challenges were presented in power efficiency.To tackle this challenge, following the idea of edge intelligence, a deep reinforcement learning approach was introduced to dynamically configure the computational scale of the deep learning model on UAV and hence realize the autonomous navigation with low latency and high energy efficiency.Evaluations based on both simulation and real prototype experiments show that the proposed approach has the less energy consumption, longer navigation trail and higher obstacle avoidance rate.

Key words: UAV, edge intelligence, deep learning, reinforcement learning, autonomous navigation

中图分类号:

TP399

林椿珉, 曾烈康, 陈旭. 边缘智能驱动的高能效无人机自主导航算法研究[J]. 物联网学报, 2021, 5(2): 87-96.

Chunmin LIN, Liekang ZENG, Xu CHEN. Research on power efficient autonomous UAV navigation algorithm: an edge intelligence driven approach[J]. Chinese Journal on Internet of Things, 2021, 5(2): 87-96.

图/表 13

图1

图2

图3

图4

图5

图6

图7

图8

图9

表1

图10

表2

表3

参考文献 31

[1]	FAESSLER M , FONTANA F , FORSTER C ,et al. Autonomous,vision-based flight and live dense 3D mapping with a quadrotor micro aerial vehicle[J]. Journal of Field Robotics, 2016,33(4): 431-450.
[2]	SCHERER S , REHDER J , ACHAR S ,et al. River mapping from a flying robot:state estimation,river detection,and obstacle mapping[J]. Autonomous Robots, 2012,33(1/2): 189-214.
[3]	PALOSSI D , GOMEZ A , DRASKOVIC S ,et al. Extending the lifetime of nano-blimps via dynamic motor control[J]. Journal of Signal Processing Systems, 2019,91(3/4): 339-361.
[4]	ZHOU Z , CHEN X , LI E ,et al. Edge intelligence:paving the last mile of artificial intelligence with edge computing[J]. Proceedings of the IEEE, 2019,107(8): 1738-1762.
[5]	SHEN S J , MULGAONKAR Y , MICHAEL N ,et al. Multi-sensor fusion for robust autonomous flight in indoor and outdoor environments with a rotorcraft MAV[C]// 2014 IEEE International Conference on Robotics and Automation (ICRA). Piscataway:IEEE Press, 2014: 4974-4981.
[6]	WEISS S , SCARAMUZZA D , SIEGWART R . Monocular-SLAMbased navigation for autonomous micro helicopters in GPS-denied environments[J]. Journal of Field Robotics, 2011,28(6): 854-874.
[7]	ROSS S , MELIK-BARKHUDAROV N , SHANKAR K S ,et al. Learning monocular reactive UAV control in cluttered natural environments[C]// 2013 IEEE International Conference on Robotics and Automation. Piscataway:IEEE Press, 2013: 1765-1772.
[8]	GIUSTI A , GUZZI J , CIRE?AN D C ,et al. A machine learning approach to visual perception of forest trails for mobile robots[J]. IEEE Robotics and Automation Letters, 2016,1(2): 661-667.
[9]	SMOLYANSKIY N , KAMENEV A , SMITH J ,et al. Toward low-flying autonomous MAV trail navigation using deep neural networks for environmental awareness[C]// 2017 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). Piscataway:IEEE Press, 2017: 4241-4247.
[10]	MEHROOZ G , EBEID E , SCHNEIDER-KAMP P . System design of an open-source cloud-based framework for Internet of drones application[C]// 2019 22nd Euromicro Conference on Digital System Design (DSD). Piscataway:IEEE Press, 2019: 572-579.
[11]	PALOSSI D , LOQUERCIO A , CONTI F ,et al. A 64-mW DNN-based visual navigation engine for autonomous nano-drones[J]. IEEE Internet of Things Journal, 2019,6(5): 8357-8371.
[12]	李肯立, 刘楚波 . 边缘智能:现状和展望[J]. 大数据, 2019,5(3): 69-75.
	LI K L , LIU C B . Edge intelligence:state-of-the-art and expectations[J]. Big Data Research, 2019,5(3): 69-75.
[13]	莫梓嘉, 高志鹏, 苗东 . 边缘智能:人工智能向边缘分布式拓展的新触角[J]. 数据与计算发展前沿, 2020,2(4): 16-27.
	MO Z J , GAO Z P , MIAO D . Edge intelligence:a new exploration for artificial intelligence expanding to edge[J]. Frontiers of Data ＆ Computing, 2020,2(4): 16-27.
[14]	ZHANG X Z , WANG Y F , LU S D ,et al. OpenEI:an open framework for edge intelligence[C]// 2019 IEEE 39th International Conference on Distributed Computing Systems (ICDCS). Piscataway:IEEE Press, 2019: 1840-1851.
[15]	FRAGKOS G , KEMP N , TSIROPOULOU E E ,et al. Artificial intelligence empowered UAVs data offloading in mobile edge computing[C]// ICC 2020-2020 IEEE International Conference on Communications (ICC). Piscataway:IEEE Press, 2020: 1-7.
[16]	张星洲, 鲁思迪, 施巍松 . 边缘智能中的协同计算技术研究[J]. 人工智能, 2019,6(5): 55-67.
	ZHANG X Z , LU S D , SHI W S . Research on collaborative computing technology in edge intelligence[J]. Artificial Intelligence, 2019,6(5): 55-67.
[17]	刘通, 方璐, 高洪皓 . 边缘计算中任务卸载研究综述[J]. 计算机科学, 2021,48(1): 11-15.
	LIU T , FANG L , GAO H H . Survey of task offloading in edge computing[J]. Computer Science, 2021,48(1): 11-15.
[18]	MCMAHAN B , RAMAGE D . Federated learning:collaborative machine learning without centralized training data[J]. Google Research Blog, 2017:3.
[19]	FANG B Y , ZENG X , ZHANG M . NestDNN:resource-aware multi-tenant on-device deep learning for continuous mobile vision[C]// MobiCom’18:Proceedings of the 24th Annual International Conference on Mobile Computing and Networking.[S.l.:s.n.], 2018: 115-127.
[20]	董超, 沈赟, 屈毓锛 . 基于无人机的边缘智能计算研究综述[J]. 智能科学与技术学报, 2020,2(3): 227-239.
	DONG C , SHEN Y , QU Y B . A survey of UAV-based edge intelligent computing[J]. Chinese Journal of Intelligent Science and Technology, 2020,2(3): 227-239.
[21]	ZISSERMAN A , SIMONYAN , KAREN ,et al. Very deep convolutional networks for large-scale image recognition[J]. arXiv:1409.1556, 2014.
[22]	SZEGEDY C , LIU W , JIA Y Q ,et al. Going deeper with convolutions[C]// 2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway:IEEE Press, 2015: 1-9.
[23]	HE K M , ZHANG X Y , REN S Q ,et al. Deep residual learning for image recognition[C]// 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway:IEEE Press, 2016: 770-778.
[24]	HE K M , ZHANG X Y , REN S Q ,et al. Spatial pyramid pooling in deep convolutional networks for visual recognition[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2015,37(9): 1904-1916.
[25]	SHAN S , DEY D , LOVETT C ,et al. Airsim:high-fidelity visual and physical simulation for autonomousvehicles[J]. Field and Service Robotics, 2018: 621-635.
[26]	YU H H , WINKLER S . Image complexity and spatial information[C]// 2013 5th International Workshop on Quality of Multimedia Experience (QoMEX). Piscataway:IEEE Press, 2013: 12-17.
[27]	PERLIO J , HYVARINEN A . Modelling image complexity by independent component analysis,with application to content-based image retrieval[J]. International Conferenceon Artificial Neural Networks, 2009: 704-714.
[28]	ROMERO J , MACHADO P , CARBALLAL A ,et al. Using complexity estimates in aesthetic image classification[J]. Journal of Mathematics and the Arts, 2012,6(2/3): 125-136.
[29]	KAZAKOVA N , MARGALA M , DURDLE N G . Sobel edge detection processor for a real-time volume rendering system[C]// 2004 IEEE International Symposium on Circuits and Systems. Piscataway:IEEE Press, 2004.
[30]	AWATE Y P , . Policy-gradient based actor-critic algorithms[C]// 2009 WRI Global Congress on Intelligent Systems. Piscataway:IEEE Press, 2009: 505-509.
[31]	PALACIN J , PALLEJA T , VALGANON I ,et al. Measuring coverage performances of a floor cleaning mobile robot using a vision system[C]// Proceedings of the 2005 IEEE International Conference on Robotics and Automation. Piscataway:IEEE Press, 2005: 4236-4241.

算法	距离/m	成功率	功率/W
静态配置	73.6	89%	6.1
阈值配置	67.2	83%	4.2
强化学习配置	77.6	92%	3.9

CPU	2.00 GHz 四核ARM Cortex-A572.00 GHz 双核Denver 2
内存	1 600 MHz 8 GB LPDDR4
GPU	Pascal架构，256个CUDA核心
	理想状态5
功耗/W	满载状态15
	平均9.5

算法	静态配置	阈值配置	强化学习配置
续航时间/s	821	879	945