机器人导航与探索：新时代交叉学科的突破

需积分: 10 198 浏览量更新于2024-07-17 收藏 14.63MB PDF 举报

《机器人建图与探索》是Springer Tracts in Advanced Robotics系列的第55卷，由布鲁诺·西西里亚诺、奥苏玛·卡提布和弗兰斯·格罗恩三位知名教授担任编辑。该书聚焦于机器人技术在新世纪的重要转型，尤其是其在建图（地图构建）和探索领域的最新进展。随着机器人技术的成熟，其应用范围不再局限于工业，而是延伸至人类生活的各个领域，如家庭、工作场所、社区，为服务、娱乐、教育、医疗、制造业和辅助服务提供支持。本书探讨了机器人技术如何跨越生物力学、触觉感知（haptics）、神经科学、虚拟模拟、动画、手术以及传感器网络等多个学科，推动了这些领域的进步，并反过来从新兴领域的需求中汲取灵感，促进了整个机器人技术的革新。特别是在不同学科交叉的前沿，如机器人与神经科学结合的脑机接口，或者机器人在复杂环境中的自主探索，这些领域的突破性成果尤为显著。作者克日利斯·斯塔赫尼斯博士来自弗莱堡阿尔伯特·路德维希大学计算机科学研究所，他在这本书中可能探讨了自主智能系统在建图和探索任务中的实现方法，包括机器人如何通过机器视觉、导航算法以及数据融合来构建动态环境的地图，并在未知环境中进行有效的搜索和路径规划。书中不仅涵盖了理论研究，还可能包含实际案例分析和实验结果，展示了如何将理论知识转化为实践应用，以提升机器人的适应性和可靠性。此外，该书的出版标志着机器人技术正在朝着更加智能化、协作化和多领域应用的方向发展，预示着未来机器人将在人类日常生活和工作中扮演越来越重要的角色。 ISBN 978-3-642-01096-5和e-ISBN 978-3-642-01097-2是该书的识别号，doi 10.1007/978-3-642-01097-2是在线获取的数字对象唯一标识符。《机器人建图与探索》是一本具有前瞻性和深度的学术著作，对于任何对机器人技术感兴趣的读者，无论是研究人员还是工程师，都是了解和探索该领域前沿知识的重要参考资料。

XVIII Contents

8.4 RelatedWork......................................... 141

8.5 Conclusion ........................................... 142

9 Information Gain-based Exploration ..................... 143

9.1 Introduction .......................................... 143

9.2 The Uncertainty of a Rao-Blackwellized Mapper . . . . . . . . . . . 145

9.3 TheExpectedInformationGain......................... 147

9.4 ComputingtheSetofActions........................... 151

9.5 ExperimentalResults .................................. 152

9.5.1 RealWorldApplication .......................... 152

9.5.2 DecisionProcess ................................ 154

9.5.3 Comparison to Previous Approaches . . . . . . . . . . . . . . . 154

9.5.4 CorridorExploration ............................ 157

9.6 RelatedWork......................................... 157

9.7 Conclusion ........................................... 160

10 Mapping and Localization in Non-Static

Environments............................................ 161

10.1 Introduction .......................................... 161

10.2 LearningMapsofLow-DynamicEnvironments ............ 163

10.2.1 MapSegmentation .............................. 163

10.2.2 Learning Conﬁgurations of the Environment . . . . . . . . 163

10.2.3 MapClustering ................................. 164

10.3 Monte-CarloLocalizationUsingPatch-Maps.............. 166

10.4 ExperimentalResults .................................. 168

10.4.1 ApplicationinanOﬃceEnvironment.............. 168

10.4.2 Localizing the Robot and Estimating the State of

theEnvironment ................................ 171

10.4.3 GlobalLocalization.............................. 173

10.5 RelatedWork......................................... 173

10.6 Conclusion ........................................... 175

11 Conclusion .............................................. 177

A Appendix ................................................ 181

A.1 Probability Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181

A.1.1 ProductRule ................................... 181

A.1.2 Independence................................... 181

A.1.3 Bayes’Rule..................................... 181

A.1.4 Marginalization ................................. 182

A.1.5 Law of Total Probability . . . . . . . . . . . . . . . . . . . . . . . . . 182

A.1.6 MarkovAssumption ............................. 182

References ................................................... 183

Index ........................................................ 195

Introduction

Models of the environment are needed for a wide range of robotic applications,

from search and rescue to automated vacuum cleaning. Learning maps has

therefore been a major research focus in the robotics community over the last

decades.

In general, learning maps with single-robot systems requires the solution

of three tasks, which are mapping, localization,andpath planning. Mapping is

the problem of integrating the information gathered with the robot’s sensors

into a given representation. It can be described by the question “What does

the world look like?” Central aspects in mapping are the representation of

the environment and the interpretation of sensor data. In contrast to this,

localization is the problem of estimating the pose of the robot relative to a

map. In other words, the robot has to answer the question, “Where am I?”

Typically, one distinguishes between pose tracking, where the initial pose of

the vehicle is known, and global localization, in which no a priori knowledge

about the starting position is given. Finally, the path planning or motion

control problem involves the question of how to eﬃciently guide a vehicle to

a desired location or along a trajectory. Expressed as a simple question, this

problem can be described as, “How can I reach a given location?”

Unfortunately, these three tasks cannot be solved independently of each

other. Before a robot can answer the question of what the environment looks

like given a set of observations, it needs to know from which locations these

observations have been made. At the same time, it is hard to estimate the

current position of a vehicle without a map. Planning a path to a goal location

is also tightly coupled with the knowledge of what the environment looks like

as well as with the information about the current pose of the robot.

The diagram in Figure 1.1 depicts the mapping, localization, and path

planning tasks as well as the combined problems in the overlapping areas.

Simultaneous localization and mapping (SLAM) is the problem of building a

map while at the same time localizing the robot within that map. One cannot

decouple both tasks and solve them independently. Therefore, SLAM is often

referred to as a chicken or egg problem: A good map is needed for localization

C. Stachniss: Robotic Mapping and Exploration, STAR 55, pp. 3–6.

springerlink.com

 Springer-Verlag Berlin Heidelberg 2009

4 1 Introduction

active

localization

integrated

approaches

mapping localization

exploration

path planning/

motion control

LAM

Fig. 1.1. Tasks that need to be solved by a robot in order to acquire accurate

models of the environment. The overlapping areas represent combinations of the

mapping, localization, and path planning tasks [94].

while an accurate pose estimate is needed to build a map. Active localization

seeks to guide the robot to locations within the map to improve the pose

estimate. In contrast to this, exploration approaches assume accurate pose

information and focus on guiding the robot eﬃciently through the environ-

ment in order to build a map. The center area of the diagram represents the

so-called integrated approaches which address mapping, localization, and path

planning simultaneously. The integrated approaches are also called solutions

to the simultaneous planning, localization, and mapping (SPLAM) problem.

A solution to the SPLAM problem enables a mobile robot to acquire sensor

data by autonomously moving through its environment while at the same

time building a map. Whenever the robot is moving, it considers actions to

improve its localization, to acquire information about unknown terrain, and

to improve its map model by revisiting areas it is uncertain about. In the

end, the robot is assumed to have learned an accurate model of the whole

environment as well as determined its own pose relative to this model.

Several researchers focus on diﬀerent aspects of these problems. This is

done using single robot systems as well as teams of robots. The use of mul-

tiple robots has several advantages over single robot systems. Cooperating

robots have the potential to accomplish a task faster than a single one. Fur-

thermore, teams of robots can be expected to be more fault-tolerant than

a single robot. However, when robots operate in teams, there is the risk of

possible interference between them. The more robots that are used in the

same environment, the more time each robot may spend on detours in order

to avoid collisions with other members of the team. In most approaches, the

performance of the team is measured in terms of the overall time needed to

learn a map. This means that the robots need to be distributed over the

environment in order to avoid redundant work and to reduce the risk of in-

terference. A team of robots makes ﬁnding eﬃcient solutions to problems

like exploration more complex, since more agents are involved and so more

decisions need to be made.

1 Introduction 5

It is worth mentioning that all these problems become even more complex

in the case where the environment changes over time. Most mapping tech-

niques assume that the environment is static and does not change over time.

This, however, is an unrealistic assumption, since most places where robots

are used are populated by humans. Changes are often caused by people walk-

ing through the environment, by open and closed doors, or even by moved

furniture. One possibility to deal with dynamic aspects is to ﬁlter them out

and to map the static objects only. More challenging, however, is the problem

of integrating the information about changes into the map and utilizing such

knowledge in other robotic applications. This can enable a mobile robot to

more eﬃciently execute its tasks. For example, one can expect a robot to

more robustly localize itself in case where it knows about the typical conﬁg-

urations of the non-static aspects in its surroundings.

In summary, the key problems in the context of map learning are the

questions of

• where to guide a robot during autonomous exploration,

• how to deal with noise in the pose estimate and in the observations,

• how to deal with the uncertainty in the robot’s world model and how to

interprete the sensor data,

• how to model changes in the environment over time, and

• how to eﬃciently coordinate a team of mobile robots.

The contributions presented in this book are solutions to diﬀerent aspects

of the map learning problem which explicitely consider these ﬁve aspects. We

present approaches to autonomous exploration that take into account the un-

certainty in the world model of the robot. We minimize this uncertainty by

reasoning about possible actions to be carried out and their expected reward.

We furthermore describe how to achieve good collaboration among a team of

robots so that they eﬃciently solve an exploration task. Our approach eﬀec-

tively distributes the robots over the environment and in this way avoids re-

dundant work and reduces the risk of interference between vehicles. As a re-

sult, the overall time needed to complete the exploration mission is reduced.

To deal with the uncertainty in the pose of a robot, we present a highly ac-

curate technique to solve the SLAM problem. Our approach maintains a joint

posterior about the trajectory of the robot and the map model. It produces

highly accurate maps in an eﬃcient and robust way. In this book, we ad-

dress step-by-step the problems in the context of map learning and integrate

diﬀerent solutions into a single system. We provide an integrated approach

that simultaneously deals with mapping, localization, and path planning. It

seeks to minimize the uncertainty in the map and trajectory estimate based on

the expected information gain of future actions. It takes into account poten-

tial observation sequences to estimate the uncertainty reduction in the world

model when carrying out a speciﬁc action. Additionally, we focus on mapping

and localization in non-static environments. Our approach allows the robot

剩余205页未读，继续阅读

蠡洋

粉丝: 6
资源: 4

机器人导航与探索：新时代交叉学科的突破

Hybrid Mapping

Robotic tool box加载测试

扫地吸尘机器人外文文献

sim2real挑战赛学习资料

robotic system toolbox查看机器人参数指令

有关柔性机械臂的假设模态法的20篇文献

帮我找到近三年有关多机器人区域覆盖问题的国外作者的文献资料

怎样在matlab robotic toolbox中初始化teach（）函数

labview-robtics

Robotic Vacuum Cleaner算法

最新资源