Creating Autonomous Vehicle Systems.pdf

创建自主车辆系统 （Creating Autonomous Vehicle Systems）2018版Creating Autonomous VehicleystemsCopyright C 2018 by Morgan Claypoolany form or by any means-electronic, mechanical, photocopy, recording, or any other except for brief quota- nAll rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmittedtions in printed reviews, without the prior permission of the publisherCreating Autonomous Vehicle SystemsShaoshan Liu, Liyun Li, Jie Tang, Shuang Wu, and Jean-Luc Gaudiotwww.morganclaypool.comISBN:9781681730073 printISBN:9781681730080 ebookISBN:9781681731674epubDOI:10.2200/S00787ED1V01Y201707CSL009A Publication in the Morgan claypool Publishers seriesSYNTHESIS LECTURES ON COMPUTER SCIENCE.#9Series issn:1932-122 8 Print 1932-1686 ElectronicCreating Autonomous VehicleSystemsShaoshan liuHerceptInLiyun liBaidu U.Sle langSouth China University of TechnologyhuangSl11uJean-Luc GaudiotUniversity of California, IrvineMORGAN& CLAYPOOL PUbLishersABSTRACTThis book is the first technical overview of autonomous vehicles written for a general computingand engineering audience. The authors share their practical experiences of creating autonomousvehicle systems. These systems are complex, consisting of three major subsystems: (1)algorithmsfor localization, perception, and planning and control; (2)client systems, such as the robotics operating system and hardware platform; and (3)the cloud platform, which includes data storagesimulation, high-definition(HD)mapping, and deep learning model training. The algorithmsubsystem extracts meaningful information from sensor raw data to understand its environmentand make decisions about its actions. The client subsystem integrates these algorithms to meetreal-time and reliability requirements. The cloud platform provides offline computing and storagecapabilities for autonomous vehicles. USing the cloud platform, we are able to test new algorithmsand update the HD map--plus, train better recognition, tracking, and decision modelsThis book consists of nine chapters. Chapter 1 provides an overview of autonomous vehiclesystems; Chapter 2 focuses on localization technologies; Chapter 3 discusses traditional techniquesused for perception; Chapter 4 discusses deep learning based techniques for perception; Chapter5 introduces the planning and control sub-system, especially prediction and routing technoloChapter 6 focuses on motion planning and feedback control of the planning and control subsys-tem; Chapter 7 introduces reinforcement learning-based planning and control; Chapter 8 delvesinto the details of client systems design; and Chapter 9 provides the details of cloud platforms forautonomous drivingThis book should be useful to students, researchers, and practitioners alike Whether you arean undergraduate or a graduate student interested in autonomous driving, you will find herein acomprehensive overview of the whole autonomous vehicle technology stack. If you are an autono-mous driving practitioner, the many practical techniques introduced in this book will be ofinterestto you. Researchers will also find plenty of references for an effective, deeper exploration of thevarious technologies.KEYWORDSautonomous driving, driverless cars, perception, vehicle localization, planning and control,autonomous driving hardware platform, autonomous driving cloud infrastructuresContentseface1 Introduction to Autonomous Driving1.1 Autonomous Driving Technologies Overview1.2 Autonomous Driving Algorithms11221.2.1 Sensing1.2.2 Perception1.2.3 Object Recognition and Tracking1.2.4 Action3 Autonomous Driving Client System356881.3.1 Robot Operating System (ROs)1.3.2 Hardware Platform101.4 Autonomous Driving Cloud Platform..,,111. 4.1 Simulation1.4.2 HD Map Production121.4.3 Deep Learning Model Training131.5 It Is Just the Beginning.132 Autonomous Vehicle localization152.1 Localization with gNss152.1.1 GNSS OVverview152. 1.2 GNSS Error Analysis162.1.3 Satellite-based Augmentation Systems172.1.4 Real-Time Kinematic and Differential GPS182.1.5Precise Point Positioning202.1.6 GNSS INS Integration212.2 Localization with LiDaR and high-Definition Maps222. 2.1 LiDAR Overview232.2.2 High-Definition Maps Overview252.2.3 Localization with LiDAR and HD Map292.3 Visual odometry.332.3.1 Stereo Visual Odometry342.3.2 Monocular Visual Odometry342.3.3 Visual Inertial Odometry352.4 Dead Reckoning and Wheel Odometry2.4.1 Wheel Encoders372.4.2 Wheel Odometry Errors382.4.3 Reduction of Wheel Odometry errors....,..392.5 Sensor Fusion412.5.1 CMU Boss for Urban Challenge4346.....,.412.5.2 Stanford Junior for Urban Challenge2.5.3 Bertha from Mercedes benz2.6 ReferencesPerception in Autonomous Driving513.1 Introduction513.2 Datasets513.3 Detection543.4 Segmentation563.5 Stereo, Optical Flow, and Scene Flow573.5.1 Stereo and Depth573.5.2 Optical Flo3.5.3 Scene Flo593.6 Tracking.....,613.7 Conclusions垂垂垂3. 8 References..644 Deep Learning in Autonomous Driving Perception694.1 Convolutional Neural Networks垂垂垂....,..694.2 Detection704.3 Semantic Segmentation4.4 Stereo and Optical Flow754.4.1 Stereo4.4.2 Optical flo774.5 Conclusion·804.6 References......815 Prediction and Routing835.1 Planning and Control overview.835.1.1 Architecture: Planning and Control in a Broader Sense83v115.1.2 Scope of Each Module: Solve the Problem with Modules855.2 Traffic Prediction885.2.1 Behavior Prediction as Classification...,895.2.2 Vehicle Trajectory Generation...935.3 Lane Level routing965.3.1 Constructing a Weighted Directed Graph for Routing975.3.2 Typical Routing Algorithms...,..995.3.3 Routing Graph Cost: Weak or Strong Routing..,,1035.4 Conclusions1045.5 References1046 Decision, Planning, and Control1076.1 Behavioral decisions1076.1.1 Markov Decision Process Approach.……1096.1.2 Scenario-based Divide and Conquer Approach1116.2 Motion Planning...1186.2.1 Vehicle Model, Road Model, and sL-Coordination System1206.2.2 Motion Planning with Path Planning and Speed Planning1216.2.3 Motion Planning with Longitudinal Planning and LateralPlanning…1286.3 Feedback control1326.3.1 Bicycle model1326.3.2 PID Control1346.4 Conclusions1356.5 References...1367 Reinforcement Learning-based Planning and Control1397.1 Introduction1397.2 Reinforcement Learning1407.2.1 Q-Learning.1437.2.2 Actor-Critic Methods1477.3 Learning-based Planning and Control in Autonomous Driving1497.3.1 Reinforcement Learning on Behavioral Decision1507.3.2 Reinforcement Learning on Planning and Control··...1517.4 Conclusions....1537.5 References153V1118 Client Systems for Autonomous Driving1558. 1 Autonomous Driving: A Complex system1558.2 Operating System for Autonomous Driving1578.2.1 ROS Overview1578.2.2 System Reliabilit1598.2.3 Performance Improvement1608.2.4 Resource Management and Securit.1618.3 Computing platform1618.3.1 Computing Platform Implementation.1628.3.2 Existing Computing solutions1628.3.3 Computer Architecture Design Exploration1638.4 References.....,,.,,.....1679 Cloud Platform for Autonomous Driving1699.1 Introduction··1699.2 Infrastructure...,1699.2.1 Distributed Computing Framework.1719.2.2 Distributed Storage1719.2.3 Heterogeneous Computing1729.3 Simulation9.3.1 BinPiperDD1749.3.2 Connecting Spark and ROS.1759.3.3 Performance1769.4 Model Training.1769.4.1 Why Use Spark1779.4.2 Training Platform Architecture1789.4.3 Heterogeneous Computing1799.5 HD Map generation1799.5.1 HD Map..1809.5.2 Map Generation in the Cloud.1819.6 Conclusions.....1829.7 References183Author biographies185