main-content

## Über dieses Buch

Dr. Lester A. Gerhardt Professor and Chairman Electrical, Computer, and Systems Engineering Rensselaer Polytechnic Institute Troy, New York 12180 This book is a collection of papers on the subject of Robotics and Artificial Intelligence. Most of the papers contained herein were presented as part of the program of the NATO Advanced Study Institute held in June 1983 at Castel vecchio Pascoli, Italy on the same subject. Attendance at this two week Institute was by invitation only, drawing people internationally representing industry, government and the academic community worldwide. Many of the people in attendance, as well as those presenting papers, are recognized leaders in the field. In addition to the formal paper presentations, there were several informal work­ shops. These included a workshop on sensing, a workshop on educational methodology in the subject area, as examples. This book is an outgrowth and direct result of that Institute and includes the papers presented as well as a few others which were stimulated by that meeting. A special note is the paper entitled "State-of-the-Art and Predictions for Artificial Intelligence and Robotics" by Dr. R. Nagel which appears in the Introduction and Overview chapter of this book. This paper was originally developed as part of a study for the United States Army performed by the National Research Council of the National Academy of Science and published as part of a report entitled "Applications of Robotics and Artificial Intelligence to Reduce Risk and Improve Effectiveness" by National Academy Press in 1983.

## Inhaltsverzeichnis

### State of the Art and Predictions for Artificial Intelligence and Robotics

The term robot conjures up a vision of a mechanical man—that is, some android as viewed in Star Wars or other science fiction movies. The industrial robot has no resemblance to these Star Wars figures. In reality, robots are largely constrained and defined by what we have so far managed to do with them.

Roger N. Nagel

### Artificial Intelligence and Robotics

Robotics is that field concerned with the connection of perception to action. Artificial Intelligence must have a central role in Robotics if the connection is to be intelligent. Artificial Intelligence addresses the crucial questions of: what knowledge is required in any aspect of thinking; how that knowledge should be represented; and how that knowledge should be used Robotics challenges AI by forcing it to deal with real objects in the real world. Techniques and representations developed for purely cognitive problems, often in toy domains, do not necessarily extend to meet the challenge.Robots combine mechanical effectors, sensors, and computers. AI has made significant contributions to each component. We review AI contributions to perception and object oriented reasoning. Object-oriented reasoning includes reasoning about space, path-planning, uncertainty, fitting, and friction. We concluded with three examples that illustrate the kinds of reasoning or problem solving abilities we would like to endow robots with.

### Robotics

Rather than attempt a comprehensive review of the state of the art in robotics, a monumental task in this rapidly moving field that encompasses so many diverse technical diciplines, I want instead to set forth a few central research topics which I believe will dominate the research community and largely occupy the attention of researchers for the rest of this century. In the course of my remarks I will cite a number of examples to illustrate the types of problems that have been, and will be, encountered in each of these research areas. But I make no claim that these examples provide a comprehensive overview of the field, or are even necessarily representative of the bulk of work currently on-going in the world today.

James S. Albus

### Methods for Choosing Actuators and Transmission Components for High Performance Manipulators

Recently, the importance of the design of high performance machines and particularly of robotic manipulators has been widely acknowledged [1]. However relatively little work has been done on the choice and design of the mechanical components for the best dynamic performance. Various authors have evaluated manipulator kinematic performance and have set guidelines for configuration of manipulator links and joints [2,3]. Others [A,5] have developed algorithms for defining the boundaries of the workspace for manipulators with arbitrarily complex kinematic configurations. A thorough study of the state of the art in the computer control of robot manipulators has been presented by Paul [6]. A number of authors have studied methods for enhancing manipulator performance using special control algorithms [7]. Hollerbach [8] has developed an efficient analytical formulation of manipulator dynamics for use in real time control. Methods for feedback control of flexible link manipulators have been presented by Book [9] and Book et al. [10]. Sunada and Dubowsky [11] have developed methods for the analysis of industrial manipulators with complex shaped flexible links. They have concluded that link and joint flexibility can have significant impact on system performance and stability.

W. P. Seering, K. A. Pasch

### Tactile Sensing For Robots

The purpose of this review and commentary is to examine the sense of touch in robots. This includes a presentation of the history of this relatively new technology, examination of the present state-of-the-art, and discussion of future challenges and opportunities in tactile sensing for automata. Special emphasis is placed on considerations relevant to industrial assembly robots.Following several general introductory remarks, we first consider some aspects of the human touch sense which would seem to bear usefully on problems in machine sensing. This includes relationships of touch to vision and problems of coping with variables of shape, pose, object identification, and texture.The next section covers the history and present state-of-the-art in automated tactile sensing. A number of aspects of automated taction are outlined, notably technical requirements like force, torque, compliance, slip, and pattern recognition. Relationships to manipulation are introduced, and discussion of materials and transducers, information processing, and pattern recognition completes the section.The final section examines outstanding problems and near-future predictions. This is divided into several distinct but related topic areas which go beyond tactile sensing per se: materials and transduction, data handling and pattern recognition, hierarchical control, grippers and manipulators, and revised manufacturing technology.Owing to the wide range of topics included here, coverage is extensive rather than intensive. The material is intended to serve as a general introduction to a relatively new discipline which couples basic understanding with practical application in many exciting and challenging ways. Ultimately one may predict that as dextrous, sensing, intelligent robots evolve, human society will be profoundly and irreversibly affected.

Leon D. Harmon

### Automatic Recognition of Low Resolution Tactile Sensing Data Using Rapid Transformation

A pattern is the description of an object that is sensed by appropriate hardware to form a useful data set, which in turn can be processed by a computer. In this paper, we concentrate on a class of patterns arising from geometrical figure approximations as encountered in robotics applications. A wide assortment of devices and systems exists today for data sensing of the environment in the robotics world. Generally speaking, we can classify them into two major classes: non-contact sensing and contact sensing. Non-contact sensing includes optical, sonic, ultrasonic, and magnetic ranging. Contact sensing is the area of tactile sensing with which this research is primarily concerned. A suitable definition of tactile sensing is expressed as continuously variable touch sensing over an array with a certain spatial resolution.

Kai-Kuang Ma, Paul P. Wang, Jack Rebman

### On Patterns and Pattern Recognition

A set theoratic model for representing patterns and pattern classes is presented. Accordingly, a pattern P is defined as a finite non-empty set of features where feature element F is a 3-tupple, <Xi,Xj,qk>. The first two components Xi and Xj of the feature tupple F are either primitive patterns or sub-patterns appearing in a given pattern, and the third component qk is a binary predicate satisfied by Xi and Xj. It is then possible to depict P as a semantic net where nodes represent the components Xi and Xj of FεP, and the directed edge from Xi to Xj represent the predicate qk.Depending on the values of Xi and Xj, it is possible to define a given complex pattern P in more than one way such that if Xi and Xj are primitives, then the representation P° is called the zero-order definition. The n-order definition of P is obtained by utilizing the sub-patterns $${X_{{{i_n}}}},{X_{{{j_n}}}} \subset {P^{{n - 1}}}$$.Different order representation of patterns lead into the notions of object-equivalence and closure of patterns. Further, with the aid of a probability function, a modeling scheme for pattern classes become possible.The concepts of null-pattern and null-relation help define simple and complex patterns, which in turn provide a path to the previous work done by the proponents of the statistical and structural approaches to the problem of pattern recognition.

A. M. Gökeri

### Segmentation of Digital Arcs

In pattern recognition and scene analysis, there are many occasions when we want to determine whether a digital arc is straight or not. For many applications such as the production of cartographic data bases, waveform analysis and perimeter estimation, it is preferable to access the curve as a set of connected straight lines or a polygonal line. Simple schemes for converting the digital curve into this form are useful.

S. H. Y. Hung, T. Kasvand

### Automatic Synthesis of Robot Programs from CAD Specifications

Industrial robots require more and more advanced programming tools. After “teaching by showing” techniques, manipulator-level programming languages for describing tasks by sequences of robot operations have emerged, and they to-day become widespread in industry. However, writting manipulators-level programs is not easy. This difficulty arises the need for higher level languages called task-level languages, for describing assembly tasks as sequences of goal spatial relationships between objects. Translating such a description into a manipulator-level program requires to solve three main problems: grasp planning, path planning, and fine motion planning. This paper surveys recent progress toward solving these problems. Task-level languages may deeply transform the way we will program robots in the future, and they contribute to better CAD/CAM integration.

Jean-Claude Latombe

### Languages for Programming Robots

Robots are already in widespread use in factories through the world, and their numbers and types are rapidly increasing. The ways in which they are used is also diversifying: the so called first generation robots were mainly used for large batch repetitive tasks, where the expense of setting up the work place to eliminate significant variation in the presentation of the work pieces could be justified by the subsequent usefulness of the installation for that particular task. The programming of the robot represented only a small effort in relation to the total effort needed in setting up the system, and only a short time in relation to the working life of the program. The emphasis in the first generation of robots was therefore on the mechanics of the manipulators.

A. P. Ambler

This paper addresses two topics which on the surface are unrelated, the use of CAD to assist robot and sensor programming, and the use of Ada as the basis for robot programming. The association between them arises from the fact that they are being combined in an experimental facility. The facility consists of an Intel iAPX 432 multiprocessing microcomputer system, a GE TN2500 camera, an ASEA RB 6 robot and a link to a VAX 11/780 off-line computer system. The facility is being used as a testbed for various robot programming and interface strategies, and to investigate the utility of object based systems as the computer foundation of manufacturing cells. Experimental verification of techniques using information extracted from CAD models to assist in robot programming and the use of Ada are important parts of the experiment.

Richard A. Volz, Anthony C. Woo, Jan D. Wolter, Trevor N. Mudge, Jerry L. Turney, David A. Gal

### New Directions in Computer-Aided Design/Computer-Aided Manufacturing (CAD/CAM) Systems

The technology of integrated computer-aided design/computer-aided manufacturing (CAD/CAM) has been developing very rapidly in the last decade, with increasingly greater impact on the way that the world’s durable goods are produced. No one can predict exactly each step that will be taken in future. But a survey of the current state of CAD/CAM and the developmental needs arising in current applications does indicate the general course the industry will take to meet those needs. Detailed examples of current applications are taken from five General Electric Company projects that illustrate the challenges encountered in trying to implement CAD/CAM systems.

S. G. Miller

### Object Representation, Identification and Positioning from Range Data

We review the types of representations (both single level and hierarchical) and the matching algorithms which have been found useful in the design and implementation of a 3-D vision system that can model, identify, and position industrial parts.

O. D. Faugeras, M. Hebert, E. Pauchon, J. Ponce

### Representing Shape

We summarize recent work by the author and his colleagues aimed at generating rich representations of shape for two- and three-dimensional objects. An implementation of the smoothed local symmetries representation of 2D shapes is described and its use for inspection and the recognition of occluded objects sketched. The implementation uses an edge finder developed by Canny that is optimal in a precisely defined sense. An extremum principle for determining 3D shape from 2D contour is described. The extremum principle determines, for example, the symmetry plane for Kanade’s skew symmetries. The extremum principle is coupled to previous work on surface interpolation in order to interpret curved shapes. The curvature patch representation of 3D shape is based on ideas of differential geometry.

### Two Multi-Processor Systems for Low-Level Real-Time Vision

Two multi-processor systems are described that have been designed to serve as pre-processors in hierarchical computer vision systems for the automatic interpretation of image sequences in real time. Both pre-processors consist of several separate subprocessors which perform various low-level vision tasks simultaneously and independently. Freely movable windows permit the computing power of the pre-processors to be concentrated in those parts of the image which are most relevant for the task at hand. The pre-processors are used in experiments involving motion control by computer vision.

Volker Graefe

### Tracking Robot Motion in Three-Dimensional Space: a Laser Scanning Approach

Robot manipulator designs, today, make use of a variety of transducers to provide joint positioning information in the feedback loops which control the robot’s movements. These sensors, including encoders, tachometers, resolvers, etc., are fairly reliable and adequate for many situations. However, as more sophisticated manipulators have been developed to meet more demanding performance specifications, joint sensors have been found to be insufficient for fully describing end effector motion. Encoders and resolvers simply cannot account for the link deformations due to large payloads or for the inadequate rigidity at the joints which together contribute to dynamic deflections at the end effector. In order to implement position control of the gripper, sensory information must be provided which defines the gripper1s position and orientation in 3-D space with respect to the inertial (or world reference) frame.

Jane Macfarlane, Max Donath

### Nonlinear Control with Hierarchy for Coordinated Operation of Robots

The coordinated operation of several robots is a challenging control problem of high practical interest where relatively little work has been done so far. In the paper, a hierarchical overall system is developed for the coordinated operation including the dynamics of the robots involved. The resulting equations are given in general form as well as in application of the nonlinear control and decoupling method. The new approach is applied to two robots (AIT Robot 1 and Volkswagen Robot G60) working on a conveyor belt with a common collision space.

E. Freund

### Machines That Walk

This paper explores the notion that the control of dynamically stable legged systems that locomote in 3-space can be decomposed into a planar part and an extra-planar part. The planar part generates the large leg and body motions that raise and lower the legs to achieve stepping, that propel the system forward, and that maintain balance. The planar part of the control system deals only with 2D dynamics. The extra-planar part of the locomotion control system suppresses motion in those degrees of freedom that would cause deviation from the plane of motion. These degrees of freedom include roll of the body, yaw of the body, and translation perpendicular to the intended direction of travel.Study of 2D one-legged hopping systems shows that the planar part of the control system may be quite simple, incorporating separate controllers for hopping height, running velocity, and body attitude. We review data from such planar systems, both simulated and physical, that illustrate the ability to hop in place, to run at a specified rate, to travel from place to place, and to leap over obstacles. Simulations of a 3D one-legged system show that the extra-planar part suppresses roll and yaw with good precision, while the planar part behaves as it did in 2D. Using the planar and extra-planar controls, the simulated 3D system traversed paths composed of straight segments.

Marc H. Raibert, H. Benjamin Brown, Seshashayee S. Murthy

### Mobile Robots

Research on mobile robots began in the late sixties with the Stanford Research Institute pionneering work. Two versions of SHAKEY, an autonomous mobile robot, were built in 1968 and 1971. The main purpose of this project was “to study processes for the real-time control of a robot system that interacts with a complex environment” < NIL 69 >. Indeed, mobile robots were and still are a very convenient and powerful support for research on Artificial Intelligence oriented Robotics. They possess the capacity to provide a variety of problems at different levels of generality and difficulty in a large domain including perception, decision-making, communication, etc. which all have to be considered within the scope of the specific constraints of robotics: on-line computing, cost considerations, operating ability and reliability.

Georges Giralt

### Sensor Programming — a New Way for Teaching a Robot Paths and Sensory Patterns Simultaneously

Up to now path-programming for robots mainly consisted of a chaining of linear and circular sections generated by some interpolation scheme with procedures for assuring smooth transients from one section to the next. In modern robots sensory-based path modifications are possible but in general it is not easy to tell the robot how to react on certain sensory patterns. Robot languages and CAD-techniques have made a lot of progress, but in our opinion they suffer from two problems: a)a robot as well as its environment is subject to variations in mechanical conditions e.g. variable loads or tolerances that destroy precomputed geometries.b)are we really able even with our human language to describe how we react on sensory information in processes like assembly or machining?

Gerhard Hirzinger

### Application of Variable Structure Systems Theory in Manipulator Control

The design of controllers for industrial manipulators and robot arms can be a very difficult task for control engineers. The dynamics of manipulators result in non-linear differential equations with coupled terms due to Coriolis and centrifugal torques, gravitational torques and reaction torques stemming from the acceleration of other joints. Compensation for the non-linearities is possible to some extent, but it requires a detailed model of the system and is complex and costly to implement. Furthermore, the non-linearities are usually time varying.

Okyay Kaynak

### Robotic Mechanics and Animal Morphology

The wealth of information on animal structure and motion could well assist the engineer in robotic design. Books and journal articles going back to the early 1800’s document systematic observations by experimental scientists, mostly biologists and zoologists. Although this literature is steeped in the jargon of specialists and sometimes presents conflicting theories for the same observed phenomenon, the engineer with some patience may glean much from articles about the essential mechanisms by which animals move their appendages for locomotion and in gathering food.

James F. Wilson

### Applications and Requirements for Industrial Robotics

The industrial robot is more important at present as a symbol of flexible automation rather than for its economic contribution. The convenience of the robot as a symbol is that populations in each country can be counted. The populations I shall present use the first definition given in Table 1, that of the British Robot Association. An important point to remember is that the total sales value of robots is still only around 2% that of machine tool sales in the USA. An article in Business Week noted that robotics is, so far, a business of comparable size to the archery equipment industry! Its potential will only be realised when robotics departs from the classic concept of a 5 or 6-degree of freedom manipulator arm and merges, as it certainly will, to become outwardly indistinguishable from many other types of production machinery. At that stage it will be more appropriate to use a definition more like the last one in the Table and the simplistic method of counting robot populations will no longer work.

P. G. Davey

### Applications of Industrial Robots

During the last few years, the industrial robot population has increased dramatically. For industrial robots do indeed offer considerable advantages if they are used at the right time, in the right place and for the right task; these considerations spanning technical as well as economic and social factors. An investigation of the automobile industry showed that 3 – 4 workers can be replaced by one industrial robot and for 7 replaced workers only one qualified technician is required for the programming and maintenance of the industrial robots.

H. J. Warnecke

### Applications of Industrial Robots. Technical and Economical Constraints

The first industrial robot was introduced in industry as a universal handling machine some twenty years ago. Already in the late sixties, industrial robots were the center of great expectations. The only Swedish manufacturer at that time was sure that more than 40 000 robots would be working in Swedish industry alone before 1980! Also in very late years we have seen forecasts of hundreds of thousands of robots being active in the world industry during this decade.

Nils Mårtensson

### Robot Applications Within General Motors

At no time in the history of our industry have we been faced with the challenges that we see today, both in terms of the competition in our industry and also in the selectivity shown by our customers. In meeting these challenges, General Motors has initiated programs in virtually every aspect of its operations.

R. C. Beecher

### The IBM European Robotic Applications Laboratory

The basic purpose of an applications laboratory is to examine assembly robotic applications prior to committing them to our production lines.The paper will describe how the laboratory was established, how it is staffed and funded. It will discuss some of the applications that have been completed and demonstrate them on video tape. The paper will also describe how we pass on the knowledge learned by describing the Design for Robotic Assembly Course which is run each month for both the Manufacturing and Development staffs.The paper will close with a summary of the main objectives of some of the leassons learned.

D. C. Teale

### The Organizational Impacts of Computer Based Manufacturing

Current trends in world competition for sales of manufactured products and the capabilities of new production systems based upon computing and information technologies are leading inevitably, if slowly, to a very different kind of factory, one that will require significant changes in corporate organization and business strategy. The range of technology discussed at this meeting, exhibited at the recent machine tool and technology shows and reported in use at a wide variety of industrial enterprises suggests that the so-called “Factory-of-the-Future” is a real and present capability. It is a factory with very different economics; essentially a dramatic reduction in the costs associated with product variety and production change. This leads to an ability to meet the requirements of a rapidly changing and increasingly diversified marketplace; but only if firms both invest in new flexible factories and also make the organizational and structural changes that will enable the firm to take advantage of the flexibility and economy of scope available in a factory that is essentially a computer system with machine tools and robots in place of printers.

Joel D. Goldhar

### Recovering from Failures: A New Challenge for Industrial Robotics

Unexpected events can cause the failure of apparently “correct” robot programs. The interaction with the real world makes errors unpredictable. The error recovery activity is usually programmed explicitly in the user program. This requires the user to be responsible for anticipating all the possible errors and to determine the actions he wants to take to recover from them. The aim of this paper is to present a general framework in which the activity of error recovery can be automated. This requires a model of the world in which the robot is operating, a way to interpret sensor data, and a method to generate recovery actions.

Maria Gini, Giuseppina Gini

### Introduction to Robotics in Medicine

The practice of medicine, through application of art and science, has been altered significantly during the past 10 to 15 years. Today’s medical practitioner must be part clinician, part technologist and part innovator. The great and complex pharmacological and technological advances in diagnosis and treatment have changed dramatically the manner in which medicine is practiced.

Richard C. Reba

### Expert Systems and the Concept of Knowledge

The paper gives an overview of the function and structure of expert systems. Although expert systems provide “decision support” they differ radically from traditional decision support techniques in the way that knowledge is represented and used. The main features of expert system knowledge representation are described, and the major methods for building knowledge bases are discussed.

John Fox

### Robotics at a Biomedical Cyclotron Facility

We work in a research group at the Brookhaven National Laboratory, and one of this group’s primary missions is that of studying normal and abnormal physiological brain functions. The physical tools for carrying on these studies include two medical cyclotrons, a radiopharmaceutical chemistry laboratory, the Positron Emission Transaxial Tomograph (PETT), and several digital computers. In our group are chemists, physicians, engineers, computer scientists, and a variety of professional technologists. Others involved in the research program are the human and baboon subjects whose brain functions we measure.

Jerome A. G. Russell, Alfred P. Wolf

### Medical Laboratory Automation Using Robotics

Sophisticated analytical technologies for the medical laboratory are developing at a rapid rate. Instrumental techniques such as gas chromatography/mass spectrometry, Fourier transform infrared spectroscopy and nuclear magnetic resonance spectrometry have extended the range and improved the sensitivity of sample attributes which can be measured. At the same time, improved assay techniques such as enzyme immunoassay, radioimmunoassay, enzyme-linked immunosorbent assays and fluorometric assays have considerably increased the number of biomedical substances (such as antibodies and drugs) which can be quantitatively measured.

Matthew L. Severns, Gerald L. Hawk

### The Application of Robotics to the Patient with High Spinal Cord Injury (Quadriplegia): The Robotic Arm/Work Table

Fast moving microprocessor technology leading to low-cost, highly capable microcomputer systems has been a key element in a research program aimed at examining the practicality of using a robot arm to assist the physically handicapped. A particularly critical need exists for the high spinal cord injured person for devices that increase self-care and independence for long periods of the day. Such individuals with total loss of upper and lower limb function represent one of the areas of greatest need in rehabilitation engineering. A research program at the Applied Physics Laboratory of The Johns Hopkins University has been underway since 1974 to examine the applicability of robotics to the solution of this problem. The program, sponsored by the Veterans Administration, has now reached the stage of clinical testing at Spinal Cord Injury Centers at VA Medical Centers in Richmond, Virginia, and Cleveland, Ohio.

W. Seamone

### Industrial Robots: Present and Future

“Robot” has become a household word today, but in 1923 when the play-write Karel Capek first saw his play R.U.R. translated into other languages, including English, the concept of a humanlike automaton which could perform work at its masters bidding had only been dealt with in mythology and folklore in terms of magic and sorcery. In the late 1930’s stories based on so-called “science-fiction” became increasingly popular and it was in 1939 that Isaac Asimov entered this burgeoning short-story field. In 1942, he postulated his “three laws of robotics” which are still probably more often quoted and recognized than anything written by a modern author. At this time, industrial and manufacturing technology were making such rapid strides that by 1956 George C. Devol’s patent for “Programmed Article Transfer” could be accepted as technically realizable and his ideas were adopted and made real by the group of engineers headed and inspired by Joe Engelberger. 1961 saw the first successful industrial robot application in the United States and before 1970 arrived it was almost commonplace to see multiple robot installations in the automotive industry.

Maurice J. Dunne

### Social Implications of Automation

The social implications of automation are many and diverse. For some people, automation has helped bring about a better standard of living, or a longer and more varied life, a more efficient use of goods and materials, and an increased understanding of the world in which we live. For other people, automation has put them out of work, it has provided them with dull, repetitive jobs, and it has prevented them from being a part of a more natural existence. By no means is there agreement that automation is uniformly good or unacceptably bad for society. On the other hand, automation through communication, computing, and control has made possible many heretofore impossible activities and events, for example, world wide color television and man’s trip to the moon.

Harold Chestnut

### Backmatter

Weitere Informationen