Delbert Tesar holds the Carol Cockrell Curran Chair in Engineering Joseph W. Geisinger is a research associate for the Robotics Research Group at The University of Texas at Austin

A Crystal Ball The age of robotics is just before us. The market for robotic technology has tripled in the United States since 1994. The opportunity to expand this market ten-fold will depend on a dramatic increase of performance while reducing costs. It can only be achieved by using the lessons learned from the personal computer industry and finding the equivalent in robotics: standardization of machine modules and a universal operating software system. This will lead to the development of dexterous manufacturing cells that are rapidly reconfigurable to do tasks such as automated warehousing, truck palletizing, food packaging, shoe manufacturing, and fettling of plastic parts. The University of Texas at Austin is providing key leadership in developing a robotic industry in the United States. This article outlines the basis for this enthusiasm for the technology and the current work of UT's Robotics Research Group. The oldest form of robotic technology was represented by automata and its first sophisticated description was given by Leonardo da Vinci about 500 years ago. This four input-four output device was intended to duplicate the complex motion of a bird's wing, perhaps 200 years before much simpler single output machines were first conceptualized. Another exceptional example was provided by J. Vaucanson in 1738, when he produced an automata to play a brief sonata with a flute (with correct fingering and air velocity control). This level of technology was transferred to complex patterns in textiles resulting in the Jacquard loom with digital inputs in the form of a continuous belt of punched cards in 1801. It was subsequently embodied in the player pianos developed during the nineteenth century. One should consider the punched cards as the stored "map" of the program to govern the operation of the system. The player piano had one input--and eighty-eight distinct and independent outputs--similar to a modern automatic screw machine used in manufacturing. Today's "electronic" map is more likely to describe the operation of an automotive fuel system in a modern automobile. This map results from carefully operated tests and experiments of the prototype system. Even though the fuel system may be extraordinarily complex, the highly refined map ensures that maximum performance is achieved under a wide range of sensed conditions. This is the modern equivalent of an intelligent machine, except that a majority of the decision making was done in advance and stored for retrieval during operation. Another recent form of this type of automata is represented by the Sarcos World Anthropomorphic figure developed by Steve Jacobson of the University of Utah. This system has a large number of degrees of freedom driven by a fixed digital memory (usually on a repeating tape). Although visually fascinating, it does not offer the significant levels of precision, speed, and force (or intelligence) required in future production systems for manufacturing. K-9, the dog in the 1960s science fiction series "Dr. Who," is, in fact, increasingly viable today. It had an encyclopedic memory and could rapidly respond to a wide range of verbal questions. Today this entertainment need is seen in the popular robots of "Star Wars." Finally, the production of numerous toy robots respond to the fascination young people have for this technology. Past Embodiments of Robotics. Early attempts to develop realistic functional systems are shown in. The manual controller is used as a master to provide kinesthetic input to the system and force feedback to the operator of a slave manipulator device (called teleoperation where the slave can be remote from the master). In this case, the master is quite human in its geometry (as is the slave manipulator). Hardyman was a prototype developed by General Electric to provide force amplification for the human, making the combination capable of picking up ten times the load than is possible by a human alone. Human augmentation remains a desirable goal, although the dominant requirement for specialized backdriveable actuators has not yet been met. The hand prototype by G. Hirzinger involves twelve degrees of freedom, twenty-eight sensors, a unique embedded actuator module, and an on-board electronic controller. This prototype is an exceptional development coming from the field of feinwerktechnik, or fine mechanics, a field common to central Europe and recently emerging in the United States as Micro-Electro-Mechanical Systems (MEMS). Developing a "walking" robot has maintained a broad interest over the past four decades. Two-legged walking prototype systems do exist but they remain far from satisfactory. The six-legged system from Ohio State University was a major effort during the 1980s. It demonstrated that six-legged walking (and some running gates) were feasible although expensive and complex. Health care has provided another forum for robot development (Figure 3). Prototype demonstrations in eye surgery have been successfully developed. Recently JPL, in concert with Dr. Steve Charles, a renown eye surgeon, designed, fabricated, and tested a miniature (six inches long) manipulator of enough resolution to be useful. Another surgical task which requires high forces and stiffness is the cutting of bone. Carnegie Mellon University re-searchers have proven this feasible using a high quality Adept industrial robot manipulator. The Adept system uses direct drive motors to improve the tracking capability necessary for this demanding physical task. Finally, Joe Engelberger, the father of American robotics, has developed a system called HelpMate. The initial use of this system is for transport in hospitals. A future use will add two manipulators (arms) to the platform to enable it to become the nurse's aid and companion to incapacitated humans. Figure 4 illustrates a range of mobile platforms to perform remote tasks. The platform in Figure 4a represents an underwater system for ocean exploration or for oil field service functions. Figure 4b shows an emerging concept by the Johnson Space Center robotics division to create an astronaut assistant. The goal is to reduce an astronaut's time outside the protected work modules by 50 percent. Finally, to survey unknown planetary surfaces, JPL is sending rovers to Mars and other planets (Figure 4c). These devices are miniaturized to reduce weight. A major research activity for the Robotics Research Group at UT Austin is shown in Figure 4. Systems of higher complexity are increasingly becoming feasible. The seventeen degrees of freedom dual arm system, now at UT, was built by Robotics Research Corporation as a prototype system for a major robot development in the mid-1980s (Figure 5a). It represents an extremely high level of dexterity and motion flexibility and is an exceptional laboratory demonstrator. The dual arm system in Figure 5b is being used to dismantle the Chicago Pile 5 at Argonne near Chicago. This is an example of a future need facing the United States and other industrialized nations: the dismantling of most of our nuclear facilities and reactors. Its basic requirement is to make it unnecessary for humans to enter a high radiation environment. Finally, Figure 5c shows a concept of a ten degrees of freedom, fifty foot-long dexterous "crane" manipulator capable of precision placement of building components with minimal human involvement, thus dramatically improving worker safety which is a major issue in this industry. One requirement here is a special light weight, high force, and high resolution actuator to drive this large structure. The market for industrial robots in the United States has tripled in the last three years, now exceeding $1 billion per year. General Motors purchases 4,000 robots per year. These systems are extraordinarily smooth with a reliability exceeding 20,000 hours (recall that cars may now be considered to be 3,000 hour machines). One of the most common applications is spot welding (Figure 6a) as well as spray painting and some assembly. The most important reality for industrial robots is that the cost to integrate (make it work) a robot into the factory is four times the cost of the robot itself. In addition, time of integration makes rapid car model changeovers virtually impossible. In order to make rapid integration feasible, it will be necessary to improve the absolute accuracy of industrial robots from 0.2 inch to 0.01 inch (a factor of 20) and to have computer control directly from the product database. Currently, no industrial robot technology is prepared to meet this dominant requirement. Another important application for robotics is electronic assembly. The Hirata manipulator (Figure 6b) uses high accuracy direct drive motors to maintain the level of speed and precision required. This Japanese-made manipulator is also used by Adept in the United States. Adept is the only U.S.-based industrial robot system manufacturer of any magnitude, leaving the enterprise open to the entry of vigorous technology based start-ups. What is Robotics? The concept of a machine equivalent to humans has always intrigued humankind and is frequently represented in various forms in the literature. It was crystallized for us by Karel Capek who coined the word robot; the Czech word robota meant the number of work days per year the serfs owed the local baron for his protection and governance. Capek wrote a sophisticated tale of the gradual rise of a robot society, the reduction of the role of humans, and the eventual genocidal destruction of the robot population to be reborn in two surviving individuals. This fascination continues today in our science fiction and in game competitions such as the recent contest between Deep Blue (of IBM) and chess master, G. Kasperov. While these manifestations are fascinating, they have little to do with reality. Think of the exceptional ability of the eye-brain combination to accurately distinguish a face among hundreds, thousands, or millions of similar "shapes" that differ only by small nuances. Consider the exceptional accuracy and fingertip control necessary by a basketball player to shoot a three-point shot. These human capabilities are perfected through trial and error perceptions and corrections obtained through rigorous training, none of which the technical field of robotics is approaching in its most aggressive form. The Deep Blue-Kasperov chess contest is not representative of the integrated multi-sensory, multi-motor responses best described as highly coupled nonlinear functions which are always in conflict to result in a refined and delicate balance. They are not simple digital (discrete) alternatives. It is the differencing (conflict resolution) that makes it possible for humans to be trained at an exceptionally high level. In fact, antagonistic control of large forces to provide a refined small force has long been known to be a difficult technical task. Yet the motion of the human eye is governed by a number of parallel acting muscles which antagonistically move the eyeball in a slewing mode at high speed and, just before focusing, changes to a highly accurate slow motion to prevent overshoot and jitter. In fact, these systems begin to fail when the antagonistic error exceeds the corrective decision making of the "analog" control system. This is a lesson of greatest importance technically. Just as measurement limits for many physical phenomena exist, similar limits exist on the control of highly nonlinear-coupled man-made systems. The human/biological system is basically analog (a continuous relationship between input command and output response) while the man-made system is increasingly digital (discrete steps in the input-output relationship). We are on the verge of a revolution in the digital control of machines. Why? Because by the year 2000, computer technology will be producing a gigaflop of computational power for $5,000. This is equivalent to three or more 1980 super computers. Hence the architectural generality and forecast of a super robot now becomes truly feasible. The fields of computer science, microelectronics, and materials science are providing support to this revolution. But the real demand on the technology comes in the field which the Japanese have called mechatronics-an intimate combination of mechanical and electrical technologies. The mechanicals must generate the physical embodiment of the system; in other words, they must create the best possible parametric representation of the system. The electricals, by means of exceptional decision making software, must resolve demand/response conflicts by fusing several hierarchical levels. In fact, as the speed of digital decision making increases, the more human (or analog) the control response will appear. Today the field of robotics is moving rapidly to a blending of these fields into a new discipline. Young people who wish to be robotics leaders will strive to excel in this emerging science of mechatronics. Industrial Robotics. The robot industry has concentrated on a monolithic design of manipulators (four to seven degrees of freedom arms) which are one-off designs in much the same way we built and operated our earliest computers. A massive lesson from computers has been learned during the last two decades on the commercial development of an open architecture for the hardware system (Dell Computers) and a generalized software for the operating system (Microsoft). These systems are so open that they can be assembled on demand and integrate virtually all technical modifications from a broad range of sources without disturbing the remainder of the system. The widespread awareness of this standardization encourages investment to occur from a variety of sources. This concentration on an open architecture enables a continuous improvement on performance while reducing cost. This is quite a contrast to the paradigm of most existing production machine technologies. It now becomes possible to open the architecture of dexterous machines or robots as well. To do this actuators (the muscles) can be produced in a small number of standard sizes to populate a wide range of systems to meet a diverse set of applications. These standardized actuators will contain sensors, motors, bearings, gear trains, brakes, electronic controller, wiring, and communication buses-a massive amount of technology. They have the same significance to machines as the electronic chip has to computers (in other words, actuators are one of the standards for investment). Perhaps seven to ten actuators in each of five distinct classes would be necessary to populate all the systems required by applications. Adding links between the actuators creates the manipulator. All that is necessary to complete the system is an open architecture system controller (now being offered by several suppliers) and a generalized operational software (under development at UT Austin) in the same format as offered for computers by Microsoft. Is this feasible? Can commercial entities make money in this manner? Yes, if they expand their markets to virtually untouched applications (food, textiles, apparel, agriculture, and so forth) that are global in nature, and much larger than those already addressed (automobiles, electronics). To do so requires establishing a fully integrated technology that is not only responsive to market demands, but reacts quickly (and at virtually no cost) to product design changes. This concept is called agile manufacturing. The application of open architecture robots must also integrate solutions to large force disturbances inherent in industrial manufacturing. The high value-added functions (drilling, routing, trimming) usually represent large force disturbances which are contained by a jig (or rigid frame). The jig maintains operational precision, but it blocks all the information flow to the central computer and is certainly not agile. Further, it can easily cost ten times more than the robot. Hence, a science of machines must be developed which makes it possible to eliminate the jig. To do so will require a whole series of new sciences (metrology, criteria fusion, performance norms) and a generalized decision making software. Some of the future applications in industry are shown in Figure 7. Many onerous repetitive physical tasks must be performed in high humidity, temperature extremes, chemical fumes, and so forth. It now becomes possible to build low cost, modular robots that can operate economically anywhere in the world. Further, nominally trained operators can replace failed robot modules (plug-and-play) and do so from a small collection of spares (just as we do now for personal computers). The robot actuator module shown in Figure 7a is representative of the modularity required. Figure 7b is a concept of an advanced micro-fab architecture that would make it possible to virtually remove the human from entering the clean room space. The inner cylindrical core would be occupied by modular robots that could be repaired by module replacement by other robots (Figure 7c). The second inner cylindrical shell would store all work (wafers) in progress. Beyond that a cylindrical shell for dedicated production machines could move to an outer cylindrical shell for major service, repair, or modification. It now is feasible to address high value added functions such as airframe assembly. Figure 7d is the nose cone of a fighter aircraft. It contains 120 parts with hundreds of rivets now assembled by hand using expensive jigs and fixtures. We are designing a finite number of link and actuator modules to be assembled, on demand, fully calibrated with integrating software to carry out this demanding assembly task. The result would be a precision assembly cell of forty plus degrees of freedom (Figure 7e). Some of the manipulators would maintain precision under load of 0.01" (at least ten times better than that available from the best industrial robot today). Some would be force robots to prevent deformation of the product. Others would be dexterous fixturing devices. The whole would be a completely reprogrammable system whose operation would be based on commands derived from the data base of the product-a true representation of agile manufacturing. UT Austin Research. The Robotics Research Group at UT Austin has a forty-year history in machine development, thirty years specifically devoted to robotics. Since 1975, much of this effort has concentrated on analytical and design infrastructure for an open (modular) architecture of systems with many degrees of freedom that can satisfy a broad range of applications for future production machines. Our overall goal is to create a standardized set of advanced actuators with a relatively low production cost. This minimal set of actuators would then create open architecture machines and manufacturing cells assembled in the same manner as are personal computers. Software sufficiently general to operate any machine assembled from these standardized machine modules is then required. Our research program has designed and developed an object-oriented software called OSCAR. This software can operate simple six degrees-of-freedom robot manipulators or complex forty degrees of freedom manufacturing cells. Its versatility allows a unified control for maximum performance, condition-based maintenance, and fault tolerance. Much of this class of technology is already operating nuclear reactors, supercomputers, and even modern automobiles. It is now possible to use this technology in future production machines. Based on this technical development, the Robotics Research Group is pursuing the following applications: Plutonium Processing. The operation of multiple robots in a glove box to handle and repackage highly radioactive plutonium. (Figure 5a) Dismantlement. The operation of sixteen degrees of freedom dual arm systems to decommission nuclear facilities and nuclear reactors. (Figure 5b) Airframe Manufacture. The development of precision manipulators to create versatile assembly cells without the use of expensive jigs and fixtures. (Figure 7d, 7e) Robonaut. The development of control software for the operation of dexterous hands and dual arm systems to assist the astronaut in space. (Figure 4b) Shipbuilding. The design and development of low cost, modular portable robots to weld ship structures at a cost/benefit ratio fifty times better than previous systems. (Figure 7a) Robot Crane. The design of a fifty-to-sixty -foot-long dexterous crane to assemble standard components of buildings with minimal human involvement thereby increasing worker safety. (Figure 5c) Summary. This is exciting business. Just the thing to attract the brightest young minds. Our goal is to move away from a simple concept of single purpose machines to those that can be assembled on demand to meet a wide range of applications at reduced costs. These systems will be fully integrated and reconfigurable, maintainable by a nominally trained technician, and repairable by module replacement from a limited number of modules kept on hand at low cost. This approach is identical to the commercially successful model for personal computers (standardized computer chips and operating systems). Dr. Delbert Tesar holds the Carol Cockrell Curran Chair in Engineering at The University of Texas at Austin. Involved in robotics since 1968, Dr. Tesar created and directs the Robotics Research Group at UT Austin, which has twenty-five graduate students working in the campus laboratory. His research interests include the development of advanced component and system technology for intelligent machines and robotics including their performance, condition-based maintenance, and fault tolerance for applications in space, manufacturing, military operations, and microsurgery. He received his Ph.D. in mechanical engineering from Georgia Institute of Technology. Dr. Tesar can be reached at 512-471-3039 or tesar@mail.utexas.edu Joseph W. Geisinger is a research associate for the Robotics Research Group at The University of Texas at Austin. His research interests include embedded servo system design, actuator control system design, and fault tolerant system analysis. Geisinger is also the chief financial officer of ARM Automation, Inc., a company he co-founded. He earned a B.S.M.E. degree from Kansas State University and a M.S.E. and Ph.D. in mechanical engineering from UT Austin. He can be reached at 512-471-3039 or jgeisinger@mail.utexas.edu