这本书涵盖了机器人领域的所有方面，以及最新进展。这对于了解机器人领域、选取研究方向有莫大的帮助。每一章都可以作为最权威的综述文章来读，因为作者都是该领域最权威的研究学者。Springer Handbooks providea concise compilation of approvedkey information on methods ofresearch. general principles andfunctional relationships in physical sciences and engineering. Theworlds leading experts in the fieldsof physics and engineering will beigned by one or several renownededitors to write the chapters comprising each volume. The contentis selected by these experts fromSIsonline content) and other systematicand approved recent publications ofphysical and technical informationdesigned to beuseful as readable desk referencebooks to give a fast and comprehensive overview and easy retrieval ofessential reliable key informationincluding tables, graphs, and bibli-graphies. References to extensivesources are providedHanprinceof roboticsBruno Siciliano, Oussama khatib(edsWith DVD-Rom, 953 Figures, 422 in four color and 84 TablesSringerEditorsProfessor bruno sicilianoPRISMA LabDipartimento di Informatica e SistemisticaUniversita degli studi di napoli federico IIVia Claudio 21, 80125 Napoli, Italysiciliano nina. itProfessor oussama KhatibArtificial Intelligence laboratoryDepartment of computer ScienceStanford UniversitStanford. CA 94305-9010 USAkhatib @cs. stanford. eduLibrary of Congress Control Number2007942155ISBN:978-3-540-23957-4e-ISBN:978-3-540-30301-5This work is subject to copyright. All rights reserved, whether the wholeor part of the material is concerned, specifically the rights of translationreprinting, reuse of illustrations, recitation, broadcasting, reproduction onmicrofilm or in any other way, and storage in data banks. Duplication ofthis publication or parts thereof is permitted only under the provisions ofthe German Copyright Law of September, 9, 1965, in its current versionand permission for use must always be obtained from Springer-VerlagViolations are liable for prosecution under the german Copyright LawSpringer is a part of Springer Science+Business Mediapringer.comC Springer-Verlag Berlin Heidelberg 2008The use of designations, trademarks, etc. in this publication does not imply,even in the absence of a specific statement, that such names are exempt fromthe relevant protective laws and regulations and therefore free for generaluseProduct liability: The publisher cannot guarantee the accuracy of anyinformation about dosage and application contained in this book. In everyindividual case the user must check such information by consulting therelevant literatureProduction and typesetting: le-tex publishing services oHG, LeipzigSenior Manager Springer Handbook: Dr. W. Skolaut, HeidelbergTypography and layout: schreiber vis, seeheimIllustrations: Hippmann GbR, SchwarzenbruckCover design: eStudio calamar Steinen barcelonaCover production: WMXDesign GmbH, HeidelbergPrinting and binding: Sturtz GmbH, wurzburgPrinted on acid free paperSPIN1091840189/3180/YL543210ForewordMy first introduction to robotics came via a phone call exoskeletal devices to enhance hu-in 1964. The caller was Fred Terman, the author of the man performance. In those daysworld-famous Radio Engineers Handbook, who was there were no microprocessors. Soat the time provost of stanford University. Dr. Terman these devices were either withoutinformed me that a computer science professor, John computer control, or tethered toMcCarthy, had just been awarded a large research grant, a remote so-called minicomputer,part of which required the development of computer- or even a mainframe computercontrolled manipulators. Someone had suggested toInitially, some in the computerTerman that it would be prudent if the mathematically science community felt that com-oriented McCarthy had some contact with mechanical puters were powerful enough todesigners. Since I was the only one on the Stanford control any mechanical device andfaculty whose specialty was mechanism design, Terman make it perform satisfactorily. We Bernard Rothdecided to phone me even though we had never met and quickly learned that this was not to Professor of MechanicalI was a young assistant professor fresh out of graduate be the case. We started on a twofoldStanford Universityschool with only 2 years at Stanfordtrack. One was to develop particuDr. Terman's phone call led me to a close association lar devices for SAIL, so that hardware demonstrationswith John McCarthy and the Stanford Artificial Intel- and proof-of-concept systems were available for theligence Laboratory(SAIL) that he founded. robotics fledgling robotics community to experiment with. Thebecame one of the pillars of my entire academic ca- other track, which was more or less moonlighted fromreer, and i have maintained my interest in teaching and the work at SAIL, was the development of a basic meresearching the subject through to the present daychanical science of robotics. I had a strong feeling thatThe modern history of robotic manipulation dates a meaningful science could be developed and that itfrom the late 1940s when servoed arms were developed would be best to think in terms of general concepts ratherin connection with master-slave manipulator systems than concentrate exclusively on particular devicesused to protect technicians handling nuclear materialsFortuitously, it turned out that the two tracks supDevelopments in this area have continued to the present ported each other very naturally and, most importantly,day. However, in the early 1960s there was very little the right students were interested in doing their reacademic or commercial activity in robotics. The first search in this area. Hardware developments proved toacademic activity was the thesis of H A. ernst, in 1961, be specific examples of more general concepts, and theat MIT. He used a slave arm equipped with touch sensors, students were able to develop both the hardware and theand ran it under computer control. The idea in his study theorywas to use the information from the touch sensors toOriginally, we purchased an arm in order to getguide the armstarted quickly. a group at Rancho Los AmigosThis was followed by the sail project and a simi- Hospital, in Los Angeles, was selling a tongue-switch-lar project started by Professor Marvin Minsky at mit, controlled motor-driven exoskeleton arm to assistwhich were the only sizeable academic ventures into patients without muscular control of their arms. We purrobotics at that time There were a few attempts at com- chased one of these and connected it to a time-sharedmercial manipulators, primarily in connection with part PDP-6 computer. The device was named Butterfingers;itproduction in the automotive industry. In the Usa there was our first experimental robot. Several films demonwere two different manipulator designs that were being strating visual feedback control, block stacking tasksexperimented with in the auto industry; one came from and obstacle avoidance were made with Butterfingers asAmerican Machine and Foundry (AMF) and the other the star performerfrom unimation IncThe first manipulator that we designed on our ownThere were also a few mechanical devices developed was known simply as the Hydraulic arm. As its nameas hand, leg, and arm prosthetics, and, a bit later, some implies, it was powered by hydraulics. The idea wasto build a very fast arm. We designed special rotary left Unimation, his colleagues Brian Carlisle and Bruceactuators, and the arm worked well. It became the ex- Shimano reorganized unimation's West Coast divisionperimental platform for testing the first ever dynamic into Adept, Inc, which to this day is the largest USanalysis and time- optimal control of a robotic arm How- manufacturer of assembly robotsever, its use was limited since the design speeds wereQuickly the modern trend of carefully detailed memuch faster than required due to the limitations of the chanical and electronic design, optimized software, andcomputational, planning, and sensing capabilities that complete system integration became the norm; to thiswere common at that timeday, this combination represents the hallmark of mostWe made an attempt to develop a truly digital arm. highly regarded robotic devices. This is the basic conThis led to a snake-like structure named the Orm( the cept behind mechatronic, a word conied in Japan asNorwegian word for snake. The Orm had several stages, a concatenation of the words mechanics and electronicseach with an array of infatable pneumatic actuators that Mechatronics that relies on computation is the essencewere either fully extended or fully contracted. The basic of the technology inherent in robotics as we know itidea was that, even though only a finite number of po- todaysitions in the workspace could be reached, these wouldAs robotics developed around the world, a largebe sufficient if there were a large number of positions. number of people started working on various aspectsA small prototype proof-of-concept Orm was developed. and specific subspecialties developed. The first big di-It led to the realization that this type of arm would not vision was between people working on manipulatorsreally serve the SAIL community.and those working on vision systems. Early on,VThe first truly functional arm from our group was sion systems seemed to hold more promise than anydesigned by victor Scheinman, who was a graduate stu- other method for giving robots information about theirdent at the time. It was the very successful Stanford arm, environmentof which over ten copies were made as research toolsThe idea was to have a television camera captureto be used in various university, government, and indus- pictures of objects in the environment, and then usetrial laboratories. The arm had six independently driven algorithms that allowed the computer images of thejoints; all driven by computer-controlled servoed, dc pictures to be analyzed, so as to infer required inforelectric motors. One joint was telescoping(prismatic) mation about location, orientation, and other propertiesand the other five were rotary(revolute)of objects. The initial successes with image systemsie whereas the geometry of Butterfingers required an were in problems dealing with positioning blocks, solv-erative solution of the inverse kinematics, the geomet- ing object manipulation problems, and reading assemblyric configuration of the Stanford Arm was chosen so drawings. It was felt that vision held potential for usethat the inverse kinematics could be programmed in any in robotic systems in connection with factory automaeasy-to-use time-efficient closed form Furthermore, the tion and space exploration This led to research intomechanical design was specifically made to be compat- software that would allow vision systems to recognizeible with the limitations inherent in timeshare computer machine parts (particularly partially occluded parts, ascontrol. Various end-effectors could be attached to act occurred in the so-called"bin-picking,problems)andas hands. On our version, the hand was in the form ragged-shaped rocksof a vise-grip jaw, with two sliding fingers driven byAfter the ability to“‘see” and move objects becamea servoed actuator(hence, a true seventh degree of free- established, the next logical need had to do with plandom). It also had a specially designed six-axis wrist force ning a sequence of events to accomplish a complex tasksensor. Victor Scheinman went on to develop other im- This led to the development of planning as an impor-portant robots: the first was a small humanoid arm with tant branch in robotics. Making fixed plans for a knownsix revolute joints. The original design was paid for by fixed environment is relatively straightforward. HowMarvin minsky at the mit ai lab. Scheinman founded ever, in robotics, one of the challenges is to let theVicar, a small company, and produced copies of this robot discover its environment, and to modify its acarm and the stanford arm for other labs. Vicarm later be- tions when the environment changes unexpectedly duecame the West Coast Division of Unimation, Inc, where to errors or unplanned events. Some early landmark studScheinman designed the PUMA manipulator under Gen- ies in this area were carried out using a vehicle namederal Motors sponsorship through Unimation Later, for Shakey, which, starting in 1966, was developed by chara company called Automatix, Scheinman developed the lie Rosen's group at the Stanford Research Institutenovel Robot World multirobot system. After Scheinman (now called SrI. Shakey had a TV camera, a triangulating range finder, bump sensors, and was connected tions were formed(the Japan Robot Association wasto DEC PDP-10 and PDP-15 computers via radio and formed in March 1971, and the robotic Industries asvideo linkssociation (Ria) was founded in 1974 in the usa) andShakey was the first mobile robot to reason about trade shows, together with application-oriented techni-its actions. It used programs that gave it the abil- cal sessions, were introduced and held on a regular basisity for independent perception world modeling. and The most important were the international symposiumaction generation. Low-level action routines took on Industrial Robots, the Conference on Industrial robotcare of simple moving, turning, and route planning. Technology(now called the International Conference onIntermediate-level actions combined the low-level ones Industrial Robot Technology), and the ria annual tradein ways that accomplished more complex tasks. The show, which is now called the International robots andhighest level programs could make and execute plans to Vision Show and Conferenceachieve high-level goals supplied by a userThe first regular series of conferences emphasizingVision is very useful for navigation, locating objects, research, rather than the industrial, aspects of roboticsand determining their relative positions and orientation. was inaugurated in 1973. It was sponsored jointly by theHowever, it is usually not sufficient for assembling parts International Center for Mechanical Sciences(CISM)or working with robots where there are environmental based in Udine, Italy, and the International Federationconstraining forces. This led to the need to measure the for the Theory of Mechanisms and Machines (IFToMM)orces and torques generated by the environment, on (Although IFToMM is still used, its meaning has beena robot and to use these measurements to control the changed to the international Federation for the pro-robot's actions. For many years, force-controlled ma- motion of Mechanism and Machine Science. )It wasnipulation became one of the main topics of study at named the Symposium on Theory and Practice of RobotsSAIL, and several other labs around the world. The use and Manipulators(RoManSy). Its trademark was anof force control in industrial practice has always lagged emphasis on the mechanical sciences and the activethe research developments in this area. This seems to participation of researchers from Eastern and Westernbe due to the fact that, while a high level of force Europe as well as north america and japan It is stillcontrol is very useful for general manipulation issues, held biannually. On a personal note, it is at RoManSspecific problems in very restricted industrial environ- where I first met each of the editors of this Handbook:ments can often be handled with limited, or no, force Dr Khatib in 1978 and Dr Siciliano in 1984. They werecontrolboth students: Bruno Siciliano had been working on hisIn the 1970s, specialized areas of study such as PhD for about one year, and Oussama Khatib had justwalking machines, hands, automated vehicles, sensor completed his PhD research. In both cases, it was loveintegration, and design for hostile environments began at first sightto develop rapidly. Today there are a large number of dif-RoManSy was quickly joined by a host of otherferent specialties studied under the heading of robotics and workshops today therSome of these specialties are classical engineering sub- number of research oriented robotics meetings that takeject areas within which results have been developed that place through the year in many countries. Currently,have been particularized to the types of machines called the largest conference is the International Conferencerobots. Examples here are kinematics, dynamics, con- on Robotics and Automation (ICRA), which regularlytrols, machine design, topology, and trajectory planning. draws well over 1000 participantsEach of these subjects has a long history predating theIn the beginning of the 1980s, the first real textstudy of robotics; yet each has been an area of in-depth book on robotic manipulation in the uSa was writtenrobotics research in order to develop its special charac- by Richard"Lou?"Paul(Richard P. Paul, Robot Manipter in regard to robotic-type systems and applications. ulators: Mathematics, Programming and Control, TheIn doing this specialized development, researchers have MIT Press, Cambridge, MA, 1981). It used the ideaenriched the classical subjects by increasing both their of taking classical subjects in mechanics and applyingcontent and scopethem to robotics. In addition there were several topAt the same time that the theory was being devel- ics developed directly from his thesis research at SAILoped, there was a parallel, although somewhat separate, (In the book, many examples are based on Scheinman'sgrowth of industrial robotics. Strong commercial devel- Stanford Arm. Pauls book was a landmark event inopment occurred in Japan and Europe, and there was the USA; it created a pattern for several influentialalso continued growth in the USA Industrial associa- future textbooks and also encouraged the creation ofspecialized robotics courses at a host of colleges and people get used to the activity being done by machinesuniversitiesand the devices get downgraded from"“ robot”to“mnaAt about this same time, new journals were created chine" Machines that do not have fixed bases, and thoseto deal primarily with research papers in the areas re- that have arm-or leg-like appendages have the advan-lated to robotics. The International Journal of robotics tage of being more likely called robots, but it is hard toResearch was founded in the spring of 1982, and three think of a consistent set of criteria that fits all the currentyears later the IEEE Journal of robotics and automation naming conventions(now the IEEE Transactions on robotics) was foundedIn actuality any machines, including familiar houseAs microprocessors became ubiquitous, the ques- hold appliances, which have microprocessors directingtion of what is or is not a robot came more into play. their actions can be considered as robots. In additionThis issue has, in my mind, never been successfully to vacuum cleaners, there are washing machines, refrig-resolved. i do not think a definition will ever be univer- erators and dishwashers that could be easily marketedsally agreed upon. There are of course the science fiction as robotic devices. There are of course a wide rangecreatures-from-outer-space varieties, and the robots of of possibilities, including those machines that havethe theater, literature, and the movies. There are ex- sensory environmental feedback and decision-makinamples of imaginary robot-like beings that predate the capabilities. In actual practice, in devices consideredindustrial revolution, but how about more down-to-Earth to be robotic, the amount of sensory and decisionrobots? In my view the definition is essentially a mov- making capability may vary from a great deal toing target that changes its character with technological noneprogress. For example, when it was first developed, In recent decades the study of robotics has expandeda ships gyro auto-compass was considered a robot To- from a discipline centered on the study of mechatronicday, it is not generally included when we list the robots devices to a much broader interdisciplinary subjectin our world. It has been demoted and is now considered An example of this is the area called human-centeredan automatic control devicerobotics. Here one deals with the interactions betweenFor many, the idea of a robot includes the concept humans and intelligent machines. This is a growing areaof multifunctionality, meaning the device is designed where the study of the interactions between robots andand built with the ability to be easily adapted or repro- humans has enlisted expertise from outside the classi-grammed to do different tasks. In theory this idea is cal robotics domain. Concepts such as emotions in bothvalid, but in practice it turns out that most robotic de- robots and people are being studied, and older areas suchvices are multifunctional in only a very limited arena. as human physiology and biology are being incorporatedIn industry it was quickly discovered that a specialized into the mainstream of robotics research These activimachine, in general, performs much better than a gen- ties enrich the field of robotics, as they introduce neweral purpose machine. furthermore, when the volume of engineering and science dimensions into the researchproduction is high enough, a specialized machine can discoursecost less to manufacture than a generalized one. So, speOriginally, the nascent robotics community was focialized robots were developed for painting, riveting, cused on getting things to work. Many early devicesquasiplanar parts assembly, press loading, circuit board were remarkable in that they worked at all, and littlestuffing, etc. In some cases robots are used in such spe- notice was taken of their limited performance. Today,cialized ways that it becomes difficult to draw the line we have sophisticated, reliable devices as part of thebetween a So-called robot and an adjustable piece of modern array of robotic systems. This progress is the"fixed" automation Much of this practical unfolding is result of the work of thousands of people throughoutcontrary to the dream of the pioneers in robotics, who the world. A lot of this work took place in universihad hoped for the development of general purpose ma- ties, government research laboratories, and companieshines that would do everything,, and hence sell in It is a tribute to the worldwide engineering and sciengreat enough volume to be relatively inexpensivetific community that it has been able to create the vastMy view is that the notion of a robot has to do with amount of information that is contained in the 64 chapwhich activities are, at a given time, associated with ters of this Handbook. Clearly these results did not arisepeople and which are associated with machines. If a ma- by any central planning or by an overall orderly schemechine suddenly becomes able to do what we normally So the editors of this handbook were faced with the dif-associate with people, the machine can be upgraded in ficult task of organizing the material into a logical andclassification and classified as a robot After a whileent wholXThe editors have accomplished this by organizing theThe third layer consists of two separate parts(a totalcontributions into a three-layer structure. The first layer of 22 chapters) that deal with advanced applications atdeals with the foundations of the subject. This layer the forefront of todays research and development. Thereconsists of a single part of nine chapters in which the are two parts to this layer one deals with field and serviceauthors lay out the root subjects: kinematics dynamics robots, and the other deals with human-centered andcontrol, mechanisms, architecture, programming, rea- lifelike robots. To the uninitiated observer, these chapterssoning, and sensing. These are the basic technological are what advanced robotics is all about. However, it isbuilding blocks for robotics study and developmentimportant to realize that many of these extraordinaryThe second layer has four parts. The first of these accomplishments would probably not exist without thedeals with robot structures; these are the arms, legs, previous developments introduced in the first two layershands, and other parts that most robots are made up of. of this HandbookAt first blush, the hardware of legs, arms, and handsIt is this intimate connection between theory andmay look quite different from each other, yet they share practice that has nurtured the growth of robotics anda common set of attributes that allows them to all be become a hallmark of modern robotics These two comtreated with the same, or closely related, aspects of the plementary aspects have been a source of great personalfundamentals described in the first layersatisfaction to those of us who have had the opportuThe second part of this layer deals with sensing nity to both research and develop robotic devices. Theand perception, which are basic abilities any truly au- contents of this handbook admirably reflect this com-tonomous robotic system must have. As was pointed out plementary aspect of the subject, and present a veryearlier, in practice, many So-called robotic devices have useful bringing together of the vast accomplishmentslittle of these abilities, but clearly the more advanced which have taken place in the last 50 years. Certainly,robots cannot exist without them, and the trend is very the contents of this Handbook will serve as a valuablemuch toward incorporating such capabilities into robotic tool and guide to those who will produce the even moredevices. The third part of this layer treats the subject ar- capable and diverse next generations of robotic deviceseas associated with the technology of manipulation and The editors and authors have my congratulations andthe interfacing of devices. The fourth part of this layer admirationis made up of eight chapters that treat mobile robots andvarious forms of distributed roboticsStanford. august 2007Bernard roth