Click for next page ( 52


The National Academies | 500 Fifth St. N.W. | Washington, D.C. 20001
Copyright © National Academy of Sciences. All rights reserved.
Terms of Use and Privacy Statement



Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.

OCR for page 51
Paper 6 ROBOTICS: THE INTERPLAY OF INDUSTRIAL AND ACADEMIC ACTIVITY OPENS A MAJOR NEW FIELD OF RESEAE~H EARLY HISTORY ffl e history of robotics, a subject area that is only now coming to the forefront of interest in computer science, is worth examining for the interplay that it exhibits between direct marketplace concerns and far-reaching research goals inspired by artificial intelligence. The intelligent robot is an old dream of mankind, robots having played a role in fiction since ancient days. The second-century Chinese general Chu-ko-Liang was reputed to have constructed robot donkeys and horses for use in his military transport operations, and in the thirteenth century, the English monk Roger Bacon was rumored to have built a talking bronze head that served him as a personal oracle. Robots are also prominent actors in twentieth-century science fiction. However, the development we shall trace first began to take on substantial flesh in the middle to late l950s. By that time, the availability of advanced servomechanism theory, the increasing sophistication and falling costs of electronics, and the fundamentally new capability provided by the stored program computer began to tempt engineers concerned with industrial automation to look at generalized, computer-controlled mechanical devices of broad potential applicability. The numerically controlled machine tool was one outcome, and the first robot manipulators another result of the climate of innovation to which the confluence of these three technological streams led. Viewed narrowly, industrial robots are simply computer-controlled machine tools specialized for the manipulation of work pieces. Seen in this light, they might be considered close relatives of numerically controlled machine tools, the most sophisticated of which are also regulated by stored programs, but which serve for cutting blank stock rather than for the manipulation and assembly of preformed parts. However, since the general environment of parts assembly is far more varied and complex than that of parts cutting, robot manipulators require programs that are more sophisticated than the simple geometric routines that suffice for numerically controlled machine tools. Hence robotics proper begins with the construction of the robot manipulators rather than with the simpler numerical machine tool technology. The loci of innovation were at first purely industrial, universities not being active in the earliest years of robotics. Two main companies, 51

OCR for page 51
52 Unimation and AMF (American Machine and Foundry) dominated developments in these years. Joseph Engelberger, George Devol, and Maurice Dunn led the early technical work at Unimation and are still active today. An aspect of the early history worth noting is that Unimation was from the start a specialized company whose future wan strongly conditioned by the need to make a success of the robot manipulators they were developing. This circumstance concentrated the attention of Unimation's management and technical developers on robotic. and did in fact lead to success. Tb survive in the relatively adverse technical environment of these pioneering days was not easy, since among other things, computers were still quite expensive and no clear market for robot devices had been established. By contrast, AMF, another pioneering company, was a large organization with much higher inertia and a much more limited focus on and commitment to robotics. In consequence of this, AMF's initial efforts soon fell by the wayside, and Rudy Malenkovic, the technically successful pioneer of robotics at AMF, moved to the Ford Motor Company, where his work concentrated more narrowly on the application of robots to automobile assembly lines. mis bit of history illustrates the critical importance of major innovations of small, fast-moving, specifically committed companies. Although Unimation survived, the years from 1956 to 1970 were lean ones for robotics. The general technological context in which robotics research was constrained to operate was insufficiently~favorable to allow any dramatic commercial success. Compared to the price of labor, the price at which robot assembly devices could be offered was simply too high. The technology was viewed with suspicion by many members of its potential customer base. Most robot-oriented companies other than Unimation simply went under or moved on to other activities. At this point, however, academic activity in robotics began to become significant. Marvin Minsky, realizing the technical depth of the robotic language/vision/manipulator control problem, persuaded AMF to give him a robot for use at the MIT artificial intelligence laboratory. m e MIT work with robots was undertaken as part of their general exploration of the broad area of artificial intelligence, and thus drew inspiration from the same very far-reaching goals as the MIT work on scene analysis, game playing, semantic nets, etc. At about the same time, John McCarthy, Jerome Feldman, and m omas Binford at Stanford began work on robotic hand-eye systems. m e most successful early outcomes of this university research lay in the areas of computer vision, geometrical reasoning and modeling, and general AI-like schemes for planning robot motions. Direct research on the manipulation problems of more immediate interest to potential industrial users of robots (e.g., the problems of grasp planning, force-guided motion control, work to close tolerances, etc.) was by contrast limited. There was, however, some significant university work on design of small, fast manipulators, especially that of Victor Scheinemann, who was associated with both Stanford and MIT before moving on to Unimation; commercial versions of the Scheinemann arm are now being offered by Unimation. Real-time software for control of the kinematic chains constituting manipulators of this type was also developed at Stanford and MIT by Roth, Piper, and Paul at Stanford and Horn at MIT based on

OCR for page 51
53 prior work of Hartenberg and Uicker. The MIT/Stanford kinematic control software was of direct practical importance, since it allowed the users of robots to plan robot motions in normal Cartesian coordinates, rather than involving them constantly with the complicated geometry of motions planned in joint-angle terms. Other significant robotic research efforts were also undertaken at the Stanford Research Institute, which concentrated on problems of locomotion and computer vision, and at Edinburgh, where problems of computer vision, but more significantly some of the basic problems of robot assembly, were also studied. The work at SRI on a robot rover that navigated in a complex room environment became well-known and helped enlarge the general view of what robotic techniques might accomplish. m e Edinburgh work on the assembly problem also demonstrated some interesting technical points, but unfortunately the Edinburgh manipulator hardware was only suitable for solving toy assembly problems that industry could not regard as realistic. This group's lack of suitable industrial and governmental contacts and sponsors provided a critical obstacle to the transfer of their ideas into serious industrial practice. It is also worth noting that some of the early university work on robots was motivated by enthusiastic expectations of immediate progress and generally reflected the hope that short intense efforts would suffice to reach goals that still have not been attained. m e Stanford hand-eye work had the goal of assemblying a Heathkit radio, which it was hoped would become possible within two years of program inception. Fifteen years later, an assembly operation of this complexity is still well beyond the state of robotic art. Since much of the early university research was funded by the Advanced Research Projects Agency of the Department of Defense, reports at that time had it that early fielding of robot troops or robot-run military vehicles was hoped for. m is, too, requires technological capabilities that we are still far from having. Nevertheless, in spite of the failure of their most ambitious expectations, these university efforts did lay a technical base and they did train initial groups of researchers that supplied manpower for the rapid growth in robotics that began during the early 1970s. THE SECOND DECADE By the early 1970s, inexpensive minicomputers were readily available, and much cheaper microcomputers were obviously on the horizon. These dramatic strides in electronic technology clearly promised to remove one significant cost factor, the cost of computing power for manipulator control, that had impeded the application of robot technology. This fact, obvious to industrial groups both in the United States and abroad, triggered an expansion in the level of industrial robotic research, with major companies such as IBM, Cincinnati Milacron, Texas Instruments, Westinghouse, GE, and GTE all becoming involved in robotic research and development. Japan became a major focus of robotic activity at this time, with Hitachi, Fujitsu, and the powerful research group at the

OCR for page 51
54 National Electronic Laboratory all building up robotic research groups. Olivetti in Italy, ASEA in Sweden, Volkswagen in Germany, and Renault in France all entered robotic research as well. While a part of this expanded activity was simply inspired by hopes that the use of robotic techniques could alleviate some of the internal productivity problems that the companies involved were facing, at least a few of the companies building up robotics groups (including Olivetti, Fujitsu, IBM, and Cincinnati Milacron) saw robotics as a potential opportunity for major expansion of existing product lines. m e level of robotic research at universities and university-related research institutions also rose substantially. The Electronics Directorate at the National Science Foundation began to fund university work in this area, and increased funding soon became available under the more specialized NSF Research Applied to National Needs (RANN) and Productivity Technology programs. The NSF funding allowed the Stanford robotics efforts to expand to three-dimensional modeling systems. A significant robotics activity was also built up at MIT's Charles Stark Draper Laboratory. The Draper work focused on the problems of industrial assembly, specifically the mating of parts that must be fitted together to very close tolerances. Here the inventiveness of a very able group of mechanical engineers contributed an outstanding mechanical device, the so-called remote center compliance device, which made it possible for a robot manipulator to mate parts that had to be fitted to closer assembly tolerances than the maximum geometric precision of the manipulators themselves. NSF-funded work at the University of Rhode Island demonstrated that computer vision could be practically and successfully combined with robot manipulation. m e Rhode Island work concentrated on a single problem of great industrial importance, namely that of picking up parts made available to a robot in disorganized tote bins. This well-chosen concentration made it possible for the Rhode Island work to succeed much more markedly than other groups whose research aims were perhaps broader but whose focus was also more diffuse. During this same period, industrial robotics pitched at a number of specialized applications began to achieve real commercial success. Concentrated attention to one specific application area--automobile spot welding--made it possible for this operation to be performed reliably and well. In this application a heavy tool (a welding gun) must be positioned to within a few hundredths of an inch, with specified orientation, at prespecified points on an automobile frame. When the tool reaches a significant point, it is activated and a weld made. m e first successful welding applications required the auto frame being welded to remain stationary, but, as computing power increased, the more complex task of making welds on a body moving along an assembly line was mastered. Other similar applications--for example, paint spraying and die casting--were also studied in depth and opened important new markets for the sale of robots. During the 1970s, various industrial and university research groups began experimenting with sensor-equipped robots. As already noted, the Stanford, SRI, and Rhode Island work involved the combination of computer vision and robot manipulation systems. Other work at

OCR for page 51
55 Stanford, IBM, and elsewhere involved the use of more or less sophisticated tactile sensors. m e use of such sensors demands control software of much greater sophistication than that needed for robots whose control is fixed and purely geometric. For fixed geometric control, simple lists defining the path that a manipulator is to traverse suffices but in the presence of sensors, one needs software that can deal with many sensed conditions, choosing among alternative actions and reacting immediately via interrupt-handling software to detected external events. Accordingly, the development of more sophisticated sensors pushed roboticists, perhaps for the first time, into active concern with the sophisticated software and programming language issues that had been central to other branches of computer science. Out of this involvement with software design questions came the control languages that are being sold with the more sophisticated of the commercial arms available today. For example, the Unimate model 250 and 500 manipulators (which are largely based upon Scheinemann's Stanford and MTT work) are programmed in a language called VAL, which is a direct derivative of the experimental AI robotic language developed during the early 1970s at Stanford. The relatively advanced AML language supporting the IBM line of robot manipulators also reflects the Stanford influence, several key members of the IBM group having been trained at Stanford. The importance of this Stanford work can be seen by contrasting the Stanford-descended manipulators with some of the other robot manipulators being sold today that still make use of more primitive software concepts that derive from pre-1970s research. For example, the Cincinnati-Milacron manipulators and some of the other robot manipulators being sold today still make use of more primitive software concepts that derive from pre-1970s research, and are programmed in a language reminiscent of the APT machine-tool programming language. The expansion of robotic research and development activity that characterized the 1970s has continued and intensified during the first years of the present decade. Westinghouse Corporation has recently established a separate division within their research and development organization to develop robots for internal use. This initiative was undertaken after a visit of the Westinghouse board chairman to Japan made him aware of the relatively advanced assembly techniques being used at Hitachi. Robots for electric motor assembly are of particular interest to Westinghouse. General Electric has announced its intention to undertake a major reorganization of its manufacturing operations, with greatly increased use of robotics. General Motors has stated that it aims to increase its population of working robots substantially. The level of university robotics-related research is also growing rapidly. A major new activity, organized as a Robotics Institute, has begun at Carnegie-Mellon University. The Carnegie-Mellon group combines the interests and talents of the university's computer science, mech- anical engineering, electrical engineering, and industrial engineering departments, and can be expected to bring a broad spectrum of pragmatic and theoretical talents, covering algorithm, software, and hardware design, to the ambitious work that they have undertaken. The University of Florida has also established a Robotics Institute, and the powerful

OCR for page 51
56 MIT group is undergoing major expansion. Many other universities are starting to involve themselves in robotics, and are actively seeking industrial connections that can ease their entry into this field. Robotics has become a matter of interest to the press and general public and is seen as a vital key to gains that will ensure the ability of the United States to compete internationally. Access to the flow of funds critical for continued robotic research and development is assured by the limited but very real practical successes outlined above. Although science fiction, perhaps abetted by the speculations of some artificial intelligence enthusiasts, has conditioned the public to expect spectacular events, incremental progress based upon complex and strenuous research efforts is more likely to characterize the coming decade. Commercially successful new applications of robotics will become possible at those points where the most advanced concepts projected by research laboratories can be cut back to yield more limited, but well-engineered, reliable devices that answer the immediate needs of industry. Industrial acceptance is, of course, critically conditioned by cost, so that robot systems limited in their kinematic and sensory complexity and in their demand for computing power will come into wide use before more flexible but expensive systems. Another factor bound to condition the rate at which industry demands robot equipment will be the need to rework the existing industrial environment of fixtures, -~ _ to better adapt them to robot-based - ~ - nart feeding, and transport devices production styles. But, as the cost of robots falls, and as these inherently universal devices grow in adaptability and reliability, broad industrial acceptance seems inevitable. CURRENT RESEARCH EMPHASES To manage the technical problems of robotics, very challenging research will have to be undertaken, and complex, expensive developmental activities mounted as well. m is will require both extensive university research efforts and major industrial developments. Here we perceive an area in which well-structured university-industry cooperation could accelerate the growth of a very challenging technology. To make a robot manipulator useful commercially involves a significant software effort, which must at the very least provide for real-time manipulator control, rapid handling of sensor-generated interrupts, and complex geometric computations. These requirements will increase significantly as multiarmed robot systems come to be employed. In robotics, computer science confronts the kinematic and dynamical reality of three-dimensional space, a circumstance that has already begun to involve robotics researchers in many fields whose relationship to the pragmatic requirements of computer science was previously marginal. Among these newly critical subject areas, computational algebra and geometry, Lagrangian dynamics, and the theory of friction and of elasticity may all be listed. All this implies that the complex of issues that researchers in robotics need to face is extraordinarily broad, so that practical progress in this field is likely to be more

OCR for page 51
57 dependent on advanced research than is the case for other computer application fields. Since little of this scientific material has until now been part of the computer science curriculum, we can also expect the requirements of robotics to encourage a substantial revision and mathematical deepening of the curriculum that university computer science departments will have to offer. Although robotics research seems certain to touch upon a par- ticularly broad range of technologies and scientific disciplines, we can gain some understanding of the areas likely to be of greatest significance over the next decades by surveying the near-term requirements of industrial robotics. These include the following: 1. In-depth studies of important current applications. Robot spotwelding has become routine, and attention is now turning to the more complex physical problems associated with continuous arc welding, where proper control of welder robots requires some understanding of the thermodynamics of the liquid-solid arc pool. Ways of specializing robots to work in environments that are hazardous or inaccessible to humans, e.g., high-purity clean rooms, deep-sea environments, nuclear reactors, and space, also require detailed study and will sometimes raise complex dynamical and other problems. 2. Improvement of visual, tactile, and other robot sensors. Current computer vision software is of limited reliability and quite expensive computationally. Much faster and more stable picture- processing algorithms and devices are required. To produce these will require penetrating theoretical research, as well as the development of specialized high-performance VLSI chips whose logic will have to embody the best algorithms that research can make available. Vision systems that are easily reprogrammable for a wide range of applications are particularly desirable, but at present it is not at all clear how these can be created. Tactile sensing plays a particularly important role in dextrous manual assembly. The subtlety of the human tactile sense is far from being matched by the relatively crude tactile sensors currently avail- able with robot manipulators. It seems clear that greatly improved sensors will be required if complex assemblies, especially of fragile and deformable parts, are to be attempted, and if more sophisticated methods of grasping are to be developed. m is has been recognized as an important research issue, and work on improved tactile sensors under way now should yield considerably improved sensors within a few years. Better sensors will in turn call for more sophisticated software to manage them, a consideration that emphasizes the importance of improved programming techniques to the general progress of robotics. It is also important to develop improved proximity sensors able to give advance warning of impending collisions. One will probably never be willing to set robot arms into rapid motion in an environment that is at all unpredictable. His makes it plain that development of better proximity sensors can contribute substantially to the productivity of robot systems, and also to their flexibility. Such systems will in turn demand software able to react appropriately to warnings that they supply.

OCR for page 51
58 3. Force-controlled motion primitives. The motion-control primitives supplied with today's robot systems are purely geometric in character, but cannot, as they stand, be used to cause a robot arm to move smoothly while maintaining contact with a curved surface of unknown shape. The ability to do this is essential for the logically flexible adaptation of a manipulator to an environment whose whole geometry is not known in precise detail. Force-controlled motions play an essential role in manual assembly, and the demonstrated advantages of devices like the Draper Laboratories' remote center compliance device point clearly to their importance for robot manipulators as well. Near-term research and development efforts to make motion primitives of this kind available in the commonly used robot-programming languages are therefore likely. 4. Improved robot-programming techniques. A large existing industrial assembly manual literature gives detailed directions for producing a great many common manufactured items. Finding some way of translating these manuals automatically into robot assembly programs would be ideal, but unfortunately, this far exceeds the capability of today's robotics programming languages. For anything close to the language of standard industrial assembly manuals to be accepted as robot control input, much more sophisticated languages than those now coming into use will be required. m e compilers for such languages will have to incorporate knowledge of the part and subpart structure of partly assembled manufactured objects. They will also have to incorporate a routine capable of planning the way in which such objects can be grasped, moved without collision through a cluttered environment, and inserted into a constrained position within a large assembly. This level of programming sophistication only becomes feasible if a robot system can either maintain a detailed model of the environment with which it is dealing through a whole complex sequence of manipulations, or acquire and refresh such a model through visual analysis of the scene before it. Although no robotic language with nearly this degree of sophistication has actually been produced, such languages have at least been projected, e.g., in the work on AUTOPASS at IBM, and its Stanford, MIT, and Edinburgh relatives AL, LAMA, and RAPT, respectively. It should be noted that the implementation of languages of this sophistication will require solution of many levels of fairly complex mathematical and geometric problems. One basic problem of this kind is that of planning collision-free motions of three-dimensional bodies through obstacle-filled environments. This problem, studied by researchers at Caltech, MIT, New York University, and elsewhere, has by now been brought to a preliminary stage of solution, but from the practical point of view this work merely reveals the complexity of computations that motion planning can involve and the importance of seeking much more efficient motion-planning schemes. The work carried out by Fahlman at MIT is also suggestive of possibilities for more advanced robot activity-planning software. Working within a simulated world of blocks, Fahlman constructed a program that could combine geometric knowledge of the collection of blocks given it with an understanding of the final assembly desired, to produce a fully sequenced assembly plan. This demonstration program was even capable of using some of the blocks available to it to construct fixtures useful in the assembly of the remaining blocks.

OCR for page 51
59 5. Improved manipulator hardware. m e essential elements of a complete robot manipulator subsystem are-a manipulator arm, the grippers and sensors with which it is furnished, and the computer circuitry that controls it. University groups can be expected to contribute substan- tially to the design of new hardware and software control schemes and improved grippers and sensors, but the high costs associated with the development of a new mechanical hardware are likely to make this a matter for industry rather than universities. It may be necessary to include features supporting advanced control concepts such as that of force-guarded motion directly in the basic mechanical structure of a manipulator, a possibility that argues for the importance of close industry-university collaboration in the continued development of robot manipulators. SOME CONCLUSIONS Advances in robotics can certainly contribute to increased U.S. productivity. However, the research issues that will need to be faced in developing robot technology are extremely broad; geometry, dynamics, elasticity theory, materials design, and electronic and software science are all involved. University researchers will find many deep issues to ponder, and industrial development groups will have many large systems to build. In order to realize the great potential of robotics, it will be particularly important to combine the abilities of these two communities and to encourage them to work closely together; this will give industry the mathematical skills that the field demands, and assist universities to find the capital resources and engineering capacity that they will need. As the potential of robot technology is realized over the next few decades through the mastery of successive practical tasks, and as the cost of robot manipulators and their controls continues to fall, economic pressures will increasingly favor wide robot deployment. The immediate technical steps that will lead in this direction are the improvement of manipulators and grippers, the close study of numerous significant applications, and the development of improved sensors and their integration into standard systems. Universities will contribute deeper investigation of the complex mathematical and programming problems of robotics. A word of caution is in order. AS the robot development works itself out, our ability to deal with its social consequences is likely to be challenged. If we fail to respond adequately to this challenge, social unrest and latter-day Luddite tendencies may become a bigger inhibition to the wide deployment of robots than technological difficulties, which in time will surely be overcome. Society will have to decide what it wants to do with the new industrial capabilities that robotics research is creating.