COMMENT ⊗   VALID 00011 PAGES
C REC  PAGE   DESCRIPTION
C00001 00001
C00002 00002	\|\\standard pox left margin, right margin requires minor change
C00005 00003
C00006 00004	\F2\CSUMMARY
C00015 00005	\F2\CBACKGROUND
C00018 00006	\CHistory
C00030 00007	\F0\CPROJECTS
C00034 00008	\JSensory Technology:  Historically, we have done considerable work in 
C00042 00009	\F2\CVISION PROJECTS
C00051 00010	\F0\CACCOMPLISHMENTS
C00081 00011	\F0\CREFERENCES      
C00091 ENDMK
C⊗;
\|\\;standard pox; left margin, right margin requires minor change
\|\\M0METM;\M1METMI;\M2METMB;\M3GRFX35;\M4FIX25;\M5NONH;
\←L\+=100;\→L\;  moving the left margin in to give the binder room
\; now set for left paging
\\←R\-=100;\→R\;  change this and line above for right paging
\; \[=4;=2;Psychologist\] underline
\←=+80;\→W\;
\←L\+W\→l\←R\-W\→r\;
\←r\+=10;\→>\←l\-=10;\→<\; 
\←l\→0\+=225;\→1\+=250;\→2\;
\⊂'10144;\←=35;\ε\;  Set for single spacing, everypage (35 space)
\←=1;\→N\;  This is the starting page number
\∞EVERYPAGE[\
\oHEAD{2[\DN]}\ This has no section number included
\\oHEAD{2[I.\DN]}\ This has the section number included.
\←R\→D\1HEAD;\¬D\→D\2HEAD;\¬D\-=1;\→D\  This is for right numbering
\\←L\→D\ This is for left numbering
\WHEAD,D=150;\
\←N\+=1;\→N]\; This increments page register by 1
\∞drawbox[\V<∧>∧t\V<∨>∨t\V<∧<∨t\V>∧>∨t]\;
\∞tocenter[\←R\-L\/=2;\→.]\;
\∞narrow[\←L\+W\→L\←R\-W\→R]\;
\∞widen[\←L\-W\→L\←R\+W\→R]\;
\∞t0[\←0\→L\→.]\;
\∞t1[\←1\+=15;\→L\→.]\;
\∞t2[\←2\+=15;\→.\→L]\;
\∞rmk[\!t1;\F1{\←.\→L]\;
\∞kmr[}\F0]\;
\∞Jrmk[\!t1;\F1{\←.\→L\J]\;
\∞kmrJ[}\F0\.]\;
\8head1(text)[\F2\C\{⊗text⊗\}=17;=3;\F0]\;USE ALL CAPS
\8head2(text)[\F2\C⊗text⊗\F0]\;USE ALL CAPS
\8head3(text)[\F1\C⊗text⊗\F0]\;Capitalize only Important Words
\8head4(text)[\F1\{⊗text⊗\}=8;=2;\F0]\;Capitalize only Important Words
\F0\;  Do not put a \, at the bottom of this page (or numbering goes wrong)




\F2\CPROPOSAL TO 



\CINTERNATIONAL BUSINESS MACHINES CORPORATION



\CAnalysis of Automated Assembly




\F2\CROBOTICS LABORATORY
\F2\CSTANFORD ARTIFICIAL INTELLIGENCE LABORATORY
\F2\CCOMPUTER SCIENCE DEPARTMENT
\F2\CSTANFORD UNIVERSITY






\CSeptember 1982








\F0\CThomas O. Binford
\CAdjunct Professor of Computer Science


\F2\CSUMMARY

\F0\JThe Robotics Laboratory of the Stanford Artificial Intelligence Laboratory
proposes to International Business Machines Corporation 
a joint effort in automated assembly using robots, for which we ask IBM to provide:
\j⊗ funding for three years at $50,000 per year;\.
\j⊗ three year loan of an RS-1 robot and an IBM 7335 robot;\.
\j⊗ a device which will be assembled;\.
\j⊗ fund any major cost for tooling for the assembly demonstration;\.
\j⊗ guidance of IBM personnel in carrying out the project.\.

\F2\CSCOPE OF WORK

\F0\JWe propose to study systems for automated assembly of precision products
in collaboration with IBM.
We plan to study the distribution of functions in assembly of one device.
We will carry out analysis, modeling of costs, and building an assembly station
for assembly of that device.  
We plan to assemble the example device using coordinated actions of two robots.
We will abstract from that example.

We propose to perform the following tasks in analysis of assembly:
\j⊗ analysis of parts handling: kitting vs separate bulk feeders or pallets;\.
\j⊗ analysis of assembly operations: single function at each station, or
all functions at one station, or an intermediate distribution of operations;\.
\j⊗ integrating a complete assembly station with two robots to assemble
one example object;  completing its assembly;\.
\j⊗ analysis of the design of one assembly from the standpoint of redesign for
automatic assembly.\.

\CPossible Separate Study

\JThe following project is not to be covered in this joint study, but may be
of interest for a separate study.  The Robotics Laboratory has strong interest
in developing generic capabilities for manipulating flexible parts, especially
wiring.\.

\F2\CTECHNICAL APPROACH

\F0\JWe plan to survey available computer programs for economic modeling,
to select one of those programs, and to carry out the analysis for an example agreed upon by 
IBM and Stanford.  We will carry out analytic study in parallel, surveying
economic models and developing models as needed.
We will maintain contact with IBM personnel and with other groups at Stanford
and SRI in carrying out this subtask in order to include the widest background 
possible in experience and knowledge relevant to the problem.

In building economic models, it is likely that a systematic or formal means
of analysis of assembly operations will be necessary.
For timing studies, a work breakdown into task components will be necessary.
We are not aware of an adequate language in use by manufacturing engineers 
for describing mechanical tasks and assembly operations.  
We have designed and implemented MSM as a system which extends MTM, a 
widely-used system for analysis of human tasks and assemblies [Mujtaba 82].  
MTM is meant for describing jobs for humans.
MSM embodies ideas for formalizing work operations for machines, especially
robots.  Operations must be specified in more detail for machines than for
people.  Particular care is required for sensor-controlled operations.
MSM, as an extension of MTM, is natural to manufacturing engineers.
We have also carried out kinematic analysis of some assembly operations
[Ohwovoriole 80].

These analyses and others will be the basis for design of a set of primitive
subtasks including sensing.  Primitive subtasks will be used in breakdown
of the assembly task example.  Parameterized models will be used for studies
of variation with parameters in system design, in order to abstract results
to a broader class of assemblies.

In building an assembly station for assembly of the example object and in
successfully completing an experimental assembly, a number of things must be
done.  Communication and synchronization must be set up between two independent
robots, the IBM RS-1 and the IBM 7535.  The two robots will be integrated with
a computer vision resident in other computers.  The entire system will be set up
as much as possible as nodes on a local network.  
Overall system software must be set up.  Tooling, special grippers, fixtures,
 and parts
feeding devices will need to be built or bought.  It is expected that IBM will
pay for any major costs of tooling for the assembly example.  We cannot predict
these costs now, but there might be two parts feeding devices at $5k-$10k each.

Stanford has extensive experience in integrating software and hardware systems,
best typified by the AL system for programming robots, an experimental
laboratory integrated around AL, and the ACRONYM system for geometric modeling
and geometric reasoning.\.

\CEDUCATION

\J
The IBM robots will be used in the education of students in robotics.
Stanford has taught a course in robotics since 1971.  It includes three
quarters of course work.  In addition, there are special research and
projects in the Robotics Laboratory.  Robotics specializations are included
in degree programs in computer science, electrical engineering, mechanical
engineering, and aeronautics and astronautics.  A special master's degree
program in artificial intelligence, CSAI, has a specialization in robotics.\.

\CREPORTING

\JAn annual report will be delivered for each year of the project.  Informal
communication will be maintained with IBM personnel at IBM San Jose and 
IBM Watson Research Lab, Yorktown, NY.\.

\CPATENTS AND LICENSING

\JTerms of patent ownership and licensing will be negotiated between Stanford
and IBM, from the starting point: a non-exclusive license to IBM for any
technology developed under this contract; patent ownership to rest with
Stanford.\.
\F2\CBACKGROUND
\CRESEARCH PROGRAM OF THE ROBOTICS LABORATORY

\F0\JThe Robotics Laboratory carries out a broad research program in computer vision
and manipulation, including:\.

\j⊗  design: design of a three finger hand; design of advanced manipulators;\.

\j⊗  dynamics and control of manipulators: 
     coupled control of a three finger hand, specified by control of the grasped
     object, not by joint motions;
     multiple, cooperating manipulators;
     end point sensing and visual tracking; adaptive control;\.

\j⊗  task level force sensory control: sensor technology; interfacing;
      analysis of force strategies; high level language for force control;
      teaching force strategies;\.

\j⊗  programming systems for programmable assembly: the AL system has an 
     interpreter, source language debugging; structured editor; information
     system, graphics for debugging and experimentation, and
     force control;\.

\j⊗   next generation of programming systems for robots:  geometric modeling 
      with geometric reasoning; simulation;
      offline programming; integrating sensing and vision;\.

\j⊗  computer-aided design: geometric modeling, simulation of kinematics
      (eventually dynamics and physics), constraint
     satisfaction, optimization, tolerancing and statistical distributions;\.

\j⊗  inspection: development of an intermediate level industrial vision system with
      less sensitivity to lighting, greater discrimination of shape, however 2D;
    discrimination of essential from cosmetic flaws;\.

\j⊗  ACRONYM integrated vision system:  ACRONYM combines geometric modeling with
      geometric reasoning in the form of a rule-based system; ACRONYM integrates
      interpretation with stereo vision and ranging, and includes edge-based segmentation.\.

\j⊗  VLSI implementation of vision operations.\.

\CHistory

\F0\JThe robotics program at Stanford  was begun by John McCarthy in 1965
at the  Artificial Intelligence Laboratory.
It now serves as the focus for most robotics activity at Stanford University.

Research in robotics at the AI Lab includes five major headings:\.
\F4  
        computer vision;
        manipulation;
        mobile robots;
        geometric modeling;
	hardware.

\F0
\JWork centers around two main systems - AL and ACRONYM.
The benefit of building these systems is that students 
can start immediately at the state of the art; they 
don't have to start at ground zero.  Projects can be continuous and large,
not limited to a single student project.  They provide an experimental system
in which to integrate results in a way that provides increasingly powerful
capability by combining modules.  The approach provides a way of testing
the generality of modules by testing them with others in new problems.
These systems provide a means of technology transfer.  An important
problem of utilizing technology is total system integration.
Our background of building systems and our current systems give us a big
advantage in technology utilization.

We have focussed on important applications to find fundamental
problems whose solutions would provide breakthroughs in these applications.
One part of the effort is problem-solving, another is problem-finding.  
In problem-solving, we find general, theoretical mechanisms to solve 
the central problems, implement these mechanisms in software and hardware,
integrate the mechanisms in systems like AL, and analyze performance.
In problem-finding, we analyze applications, analyze robotics capabilities,
identify fundamental problems, formalize these problems, and explore
theoretical and practical performance limitations.

Thus, there is a combined focus on implementation and theory.
At the Stanford Artificial Intelligence Laboratory, 
robotics is a large program with about 25 people involved on a continuous basis
and more for short projects.  There are several other groups at Stanford 
totalling perhaps 20 people.

There are many facilities at Stanford which relate to this effort which include:

\j⊗ the Center for Automation and Manufacturing Science (CAMS)
an interdisciplinary activity bringing together computer science, electrical
engineering, mechanical engineering, aeronautics and astronautics,
applied physics, and materials science;\.

\j⊗ the Center for Integrated Systems:
a major VLSI center focussing on technology transfer and user applications;
it has fabrication facilities and will build up fast turnaround fab facilities
and E-beam fabrication; a strong focus on design.\.
\F0\CPROJECTS


\JManipulator design:  We have collaborated for 15 years with
Prof Bernard Roth, Mechanical Engineering Dept.  We now
collaborate  with Carl Ruoff, JPL,
on design and control of a three-finger hand with nine degrees of freedom,
a "universal gripper".  The
motivation for the hand is to provide fine control of forces in assembly
and tool using, and to provide the relevant degrees of freedom for repositioning
objects after grasping for subsequent assembly operations. 
This is a research effort;
we expect that further effort will be necessary to make a 
rugged and reliable hand that can be used in industrial work.
Conceptual design including theoretical work, and detailed design are complete.
The fabrication of the hand is complete.  A controller is under construction.
Elegant theoretical analysis for control of the hand has been accomplished.
Work in optimizing the design of the hand has resulted in 
significant theoretical contributions [Salisbury 81] and form the basis of
the forthcoming PhD thesis of K. Salisbury.
Future work will include collaboration with the MIT AI Laboratory.

Dynamics and control:  We collaborate with the control group of the
Department of Aeronautics and Astronautics (Prof Robert Cannon and
Prof Dan Debra, with some consulting by Prof Art Bryson) on a number of important 
problems in dynamics and control of robots.  Impressive results have been
obtained in high speed slewing of flexible manipulators.
We are investigating
problems of end point sensing, tracking, load estimation, flexibility, making use of
contact sensing, closed loop kinematic chains including two cooperating manipulators 
and three finger hands; and
non-collocated sensor and actuators.
These efforts bring modern control theory to bear on deep problems which come
from making efficient use of robots by running them at high speeds rather than
degrading their performance by slowing them down, and problems which arise in
using sensing.\.

\,
\JSensory Technology:  Historically, we have done considerable work in 
technology for force and contact sensing [Binford 72]. Our 
Scheinman force sensing wrist [Scheinman 74] is one of the first
successful wrists to measure force-torque to high accuracy.
The wrist has been a basic component in industrial automation studies
involving two manipulators working in synchrony.
We are beginning new studies on contact sensors using piezoelectric
polymers and array techniques.   A project is underway to make capacitive
integrated circuit contact sensors.  We also plan to make integrated circuit
versions of sensor interfaces for strain gage sensors for the Scheinman
force sensing wrist, using capabilities of Stanford's Integrated Circuits lab.
We are currently designing force-sensing
fingers to complement force sensing wrists, to provide greater sensitivity
to small forces, and to minimize the mass between sensor and load to minimize
inertial forces caused by accelerations.\.


\JForce control and programmable compliance:  Stanford has completed a 
general and powerful analytic method for force control of robots
providing programmable compliance for assembly operations.  The mechanism
has been implemented in the AL system and has been tested in experiments
over several years.  Studies are underway on fundamental theoretical
limits to force control in exerting small, controlled forces; these studies 
will influence control and strategies
for force control, and will lead to studies of design of manipulators for
sensitive force control.\.

\JProgramming Systems for robots:  Stanford completed the AL system for
programming robots.  Currently, we are completing a portable version of
AL in PASCAL for a standalone PDP11 in the RSX11 operating system.
A small subset of AL will also be made available for small computer configurations.
The full system will include an interpreter and source language debugger
with force control, multiple device control, and software servos.
AL currently has all these capabilities.  
Plans call for release late in the year.
An AL subset has been implemented at Karlsruhe, Germany, and a language closely
patterned after AL has been implemented at the University of Tokyo by Prof Inoue.
A simulator for mechanical systems has been implemented [Soroka 80].  It will
be extended to include dynamics and control.  The simulator is written in
ACRONYM and simulates AL language programs.

Studies are underway on analysis of assembly from two standpoints: the first
is to measure times involved in fragments of assembly steps to quantify
systems issues in speeding up assemblies; the second effort is a profound
kinematic analysis of assembly from the standpoint of screw theory 
[Ohwovoriole 80] led by Prof Roth.

Work has begun on the next generation of robot programming systems, based
on a study of requirements for: a hierarchy of languages; vision; geometric
modeling; off-line programming; simulation; and planning.  The study will 
poll industry requirements.  ACRONYM provides a good basis for a new design.

Mobile robots: In collaboration with
Unimation, Inc, we will continue a long tradition of  research in mobile robots [Moravec 80].
 A new vehicle has
been built.  Stanford will do the navigation, control, and perception.
Stanford has demonstrated navigation of a robot cart in semi-realistic
environments, mapping its world by stereo perception, and planning paths
around obstacles determined by vision.

A current proposal of Prof Roth to NSF involves a basic study into the
allocation of degrees of freedom in robotic systems.  Also, he expects
to initiate a new study on universal terminal devices for manipulators.\.

\,
\F2\CVISION PROJECTS

\F0\JIntermediate Level Vision System:  We think that current industrial
vision systems have weak capabilities that limit the number of applications
that they can accomplish, and which necessitate extensive engineering for
those applications for which they are adequate. 
Industrial vision systems use thresholding to get binary pictures, and use
simple shape descriptors for two-dimensional images and models.

VLSI implementation of vision operators:  A single element of an array
of processors has been fabricated, tested, and found to be correct.  The
next step is to make a 10x10 array of processors and then a 100x100 array,
using a fault-tolerant design [Miller 81].  This processor is not the final answer to
high speed vision, but it is one part of the learning process.  Two phases
are essential: the first is architecture of vision operations; the second is
VLSI implementation of well-defined operations.

ACRONYM:  A powerful system combines geometric modeling 
with state-of-the-art problem-solving in the form of geometric reasoning,
and integrated low level visual operations [Brooks 81]. 
 ACRONYM is a three dimensional vision system.  It
is being converted to a VAX in a LISP dialect, and will be included in the
Image Understanding Program Testbed.  It will be made available on license
in about 18  months and  is expected  to be  useful for  research.  It  is
already  used by one company  on a DOD contract, and by SRI for the ARPA-DMA
Testbed.  Two universities have licensed ACRONYM.
The system will be pushed forward extensively in the next few years.
It is the underlying system for a sophisticated stereo effort for Defense
Mapping Agency, for
programmable inspection, for planning in programming robots, for the next 
generation of programming languages,
and more.  For example, a program for picking grasping points of objects was
written as a rule base in ACRONYM.  All of the work below is or will be integrated
as a part of ACRONYM.  ACRONYM is beginning to be applied to one of the
examples of the GM bin of parts series.  Our aim is not to duplicate work
done at the
University of Rhode Island on bin-picking for simple parts, but to demonstrate
a general model-based system on more complex parts.\.

\JEdge finding:  Work is underway to develop high performance edge finders
involving simplification of the Binford-Horn edge finder, and with
sophistication to provide higher performance still.  Impressive results have
been obtained and rapid progress is being made.  These techniques will be
reduced to VLSI by first simplifying the architecture and then implementing
in hardware.

Stereo vision:  A proposal has been funded to carry out one part of a program
to provide the Defense Mapping Agency with a feasibility demonstration system
for automated stereo cartography.  Stanford has made extensive progress in
automated stereo; we are in close contact with other organizations doing stereo.
We are evaluating their results and analyzing various contributions.  
Stanford has developed three previous stereo systems.
Our progress is especially rapid now in stereo vision.

Structured Light:  We are planning to resume work in structured light,
based on past work.  Stanford was one of the originators of the technique in 1970,
along with ETL, Tokyo, and IBM.  

Inferring Surfaces from Images:  Theoretical work of about three years ago
provides the outline of an incomplete theory for understanding images of
quite general scenes [Binford 81].  The results are surprisingly powerful and general.
Some results have been implemented in determining a three dimensional model
for an aircraft from a single image, using shadows [Lowe 81].  These results will be
extended greatly.  They provide the basis for effective stereo, and for
three dimensional industrial vision and inspection systems.\.

\,
\F0\CACCOMPLISHMENTS

\F0\JRobotics research has led to products.  The Scheinman arm led to
the MIT arm by Scheinman.
Several dozens of these arms were built and used by various research groups.
It has been one of the major research arms.  These developments
led to the arm produced internally by
Texas Instruments, and led to the Unimate PUMA.  The VAL system
was based on our second software system, WAVE, and was programmed by
one of our graduate students, Bruce Shimano, while he was working on
his PhD degree.  It appears widely recognized that Unimation got a 
lot out of the PUMA development.  It is not so widely recognized that
the VAL software system was also important.

The Scheinman wrist force sensor was produced for three research laboratories
other than Stanford (JPL, GM, NBS).  The design is the basis for several recent
sensor developments, including one by Unimation which may lead to a product.

The basic kinematic classification
of manipulators, solvability of manipulator configurations, 
the use of Denavit and Hartenberg matrices, modeling of
the manipulator as a closed kinematic chain, and the first
obstacle avoidance system were introduced by Pieper and Roth.
They are now standard tools.  The first work on dynamic modeling and optimal
control of manipulators was due to Kahn and Roth.
Roth showed how to optimize the kinematic structure of manipulators,
and Shimano and Roth showed how to use line geometry to define optimal structures
of manipulators.\.

\JBinford originated "structured light", independently 
along with ETL, Tokyo, with contributions from IBM [Binford 71].
Structured light has
broad importance as the basis for practical ranging devices.  Many uses of it
are under development.

[Binford 71] originated the "generalized cylinder" representation for object shape.
[Agin 72] used it in describing object primitives using structured light.
[Nevatia 74] demonstrated recognition of dolls, toy horses, and tools, using
range data.  The system learned by showing in three dimensions.
A model was put on the table in front of the laser. 
The system built its own models of objects in terms of part/whole graphs
of generalized cylinders.  The system recognized objects when shown in a
different position and orientation, with changed articulation of the object,
and with some obscuration.  The system demonstrated an indexing capability
to determine a subset of similar objects, for example, matching a doll to 
the class of quadripeds.  Indexing was based on stick figures 
using an attachment hierarchy as a coarse description.
This work is still a landmark.

[Paul 72] developed the WAVE programming system for robots (actually our second
system), and demonstrated
the first computer-controlled assembly. 
It used tools, sensing, and vision.
 A number of assemblies were carried
out in that era, including several sub-assemblies of a two-stroke gasoline engine.
It was clear that the complete assembly of the chain saw motor would be more of
the same.  Emphasis was shifted to the two crucial areas of utilizing a robot
in assembly and related tasks, namely sensorimotor control and programming.
AL was designed in 1974 and built in 1975-76.  An AL interpreter, initially
called POINTY, was developed for interactive programming.
Two other groups have developed version of AL for their own use.  AL will be
available under license in 1982.

Force control was used in a primitive way in WAVE.
[Shimano 78] developed a force control system with considerable capability.
[Salisbury 80] developed a sophisticated system for programmable compliance
and force control which appears to have sufficient speed,
generality, and sensitivity.  This is a good candidate for applications.

ACRONYM is a major advance in three dimensional vision systems [Brooks 81].
It promises to make a substantial contribution for inspection and vision
applications.  It will be available as a research system under license in 
the 1983 time period.

Major advances in inferring surfaces from images may provide significant
capability for bin picking and other three dimensional problems [Binford 81].

Recent advances in edge finding provide a basis for industrial vision systems
based on gray scale.  Advances in stereo may provide an alternative to
structured light for ranging.  The technology appears near the point of
utilization for the Defense Mapping Agency.

A line of development led to autonomous navigation by a
computer-controlled vehicle which used stereo vision to determine
a course around obstacles in a fairly realistic environment, very
slowly.

Based on recent advances, powerful computer vision is much closer to
realization than might have been imagined.  In part this follows from
finding a uniform structure for intermediate level vision operations which
appears implementable in VLSI, in part from progress on stereo, edge finding,
and grouping edges, in part because of theoretical work on interpretation.

Ohwovoriole and Roth showed how to analyze the forces involved in the general 
peg-and-hole assembly, and using screw theory, developed a general
intellectual framework for all parts mating problems.

Leifer, Roth, et al. developed a voice controlled system for rehabilitation
use, currently being tested by the Veterans Administration.\.

\JA specimen of the work of the Guidance and Control Laboratory is the
drag-free satellite.  Over a period of years, Stanford conceived the design,
built, launched, and operated the spacecraft, measuring its performance
and verifying its exceptional precision.  The design was put into use
as the Transit satellite system for ultra-precision fixing of inertial
guidance systems for ballistic missile submarines.  The development
featured sensing systems of almost unbelievable precision.  A follow-on
is application of the satellite concepts in an experiment to test
General Relativity.

Several companies have spun off from this and related projects.
Vicarm, Inc. became Unimation West.  Imagen was started to exploit
graphics printing capabilities.  Many of our students went to SRI
(seven PhDs before a group left to form the AI Center at Fairchild for Schlumberger).
They strongly influenced the work of the SRI industrial automation groups
and vision groups, especially the vision module, which among other things
led to Machine Intelligence Corporation.\.

\CPeople

\JOne of our most important products has been people who have worked in 
robotics and who are now actively contributing in the field of robotics.
A partial list follows:\.
	Donald Pieper - chief engineer, Automatix
	Gerald Gleason - chief engineer, Machine Intelligence Corp
	Jerome Feldman - former chairman, Computer Science, Univ of Rochester
	Raj Reddy - head, Robotics Institute, Carnegie-Mellon Univ
	Gerald Agin - Prof, bio-engineering, Carnegie-Mellon University
	Brian Carlisle - President, Unimation West
	Charles Spaulding - engineer, Unimation West
	Bruce Shimano - Unimation West
	Richard "Lou" Paul - Prof, Purdue University
	Kenneth Waldron - Prof, Ohio State University
	Lung-Wen Tsai - GM Research Labs
	Warren Seering - Asst Prof, MIT
	Barry Soroka - Asst Prof, USC
        J.M.Tenenbaum - Schlumberger
	Lester Earnest - President, Imagen
        R.F. Sproull - Asst Prof, CMU
        R.F. Finkel - Assoc Prof, Univ of Wisconsin
	Anthony Sword - SRI
	Bob Bolles - SRI
	Tom Garvey - SRI
	Rod Brooks - MIT
	Hans Moravec - CMU
        A.Gill - Technion, Israel
        G. Falk - Prof, Rutgers
	Don Gennery - JPL
	Patrick MacVicar-Whelan - Boeing
	Walton Perkins - GM
	Karl Pingle - ESL
	David Grossman - IBM (one year visit)
	Bruce Bullock - Hughes (one year visit)
	Maria Gini - Electrotecnico, Milan, Italy
	Giusepina Gini - Electrotecnico, Milan, Italy
	Michael Kelly - Prof, Georgia Tech
	Irwin Sobel - Columbia Univ
	Ruzena Bajcsy - Prof, Univ of Pennsylvania
	Lynn Quam - Schlumberger
        R.A.Schmidt, Jr -
	Ram Nevatia - Prof, USC
        A. Kay - Xerox
        C.R.Lui - Prof, Purdue Univ
	Bruce Baumgart - consultant
	Marsha Jo Hannah - Lockheed
	Russell Taylor - IBM
	Michael Roderick - Teledyne
	Arthur Thomas - ADAC 
        S. Ganapathy - Prof, Univ of Michigan
	William Gilbert - Texas Instruments
	Casey Cox - Automatix

\F2\CFACILITIES

\F0\JThe Robotics group of the Stanford Artificial Intelligence group
has substantial facilities including a large VAX 11/780 system
and shares facilities of the Computer Science department, including
SAIL and SCORE time-sharing systems, and the Ethernet.  Details follow.
In robotics, researchers have a large base system on which to build.
A background of working robotics laboratory systems for a decade means that
there
is a large background of worked examples which students can use and modify.

Facilities for integrated circuits in the Center for Integrated Systems are
outstanding.  Design and fabrication facilities are already excellent.
Fast-turnaround fab facilities will be available in the near
future. Facilities for sensor development in the Integrated Circuits Lab and in 
Applied Physics are excellent. 
The Guidance and Control Laboratory has fine facilities.

It is often overlooked, but the software environment
is as important as equipment.  The style of using the hardware
and software environment is also important; Stanford benefits from a decade
of developing and using top-quality systems, and from being in the Silicon
Valley, a focus of high technology development. 

A component of the proposed program is to augment the experimental laboratories
and technical support (using student staff)
where necessary to accomodate increased use of facilities to support education
and training.  It is important to utilize time of students and visiting staff
effectively by providing adequate facilities.\.


\C\F2VISION AND ROBOTICS EXPERIMENTAL SYSTEM\F0

\JTwo Unimate 600 (PUMA) arms and 
a pair of Stanford Scheinman arms are used for experiments in assembly.
The Stanford arms were designed and built here, the first in 1970, the second
in 1972. One of the PUMA arms has a servo-controlled hand and wrist force sensor.
The other will have a similar hand and force sensor soon.
One Stanford arm has a wrist force sensor which measures three components
of force and three components of torque.  It measures 300gm weight to within 5 gms.
There is a simple program which automatically  calibrates the wrist.
Two pairs of force sensing fingers have been
built, the first measures one component of force, the other measures three components.
Their sensitivity is .1 gm or so.  An IBM RS-1 robot and IBM 7335 robot are
expected in 1982.  A GCA robot built by Dainichi Kiko is expected in 1982.\.

\JThe arms are located in an assembly workstation which has a flexible
and accurate setup capability,
with accurate locating holes and screw holes for clamps.  A computer controlled
vise (with position feedback) and screwdriver (with velocity feedback) are used 
for assemblies.  More assembly tools are  planned, including a moving belt.\.

\JThe group now uses the SAIL computer system primarily (see below).  A VAX 11/780 
has just been delivered.
It has 8 M bytes of memory, two RP07 fixed media disks with 512 Mbytes
of storage each, an REM-03 67 Mbyte disk with removable media, and a TU-78
tape drive with 6250 bpi.
A PDP11/45 with 124k wds of memory controls the arms.  It is interfaced
to the KL10 I/O bus and the Ethernet.
A PDP11/60 is being interfaced for real-time arm control, with the PDP11/45
running the AL system.  The group has one SUN workstation (see below).  It will acquire
others and share other SUN workstations with the computer science department.
The group will share LISP machines with the department, but the number does
not appear to be adequate.
It has one Imagen laser printer with capability for arbitrary font and
book-quality print.\.

\JA pair of solid state cameras, GE TN2500, are used for stereo vision
and other vision at the robotics workstation.
There is also a Fairchild CCD camera with resolution 380x480.
A vision module, Machine Intelligence VS-100, is connected to a
GE TN2500 solid state camera together with a light  table.  It analyzes
silhouettes of objects for classification and location, and is coupled
to the AL system.
There is an Optronics C4500 digitizer for digitizing color photographs.\.

\JA Grinnell GMR-270 display system with 512x512 resolution with 36 bits per 
pixel provides gray scale and graphics.  It 
is shared among users by a video switch.  The system accomodates various combinations
of users: four users with 9 bit black and white
images, two users with two 9 bit channels each, or one user with 3 
color channels of 9 bits each plus
another user with a black and white channel are some of the possible configurations.
The display system has an arithmetic processor and tv digitizer.\.

\JThe robotics system will be a distributed system which communicates
over an Ethernet [Metcalfe 76].  There will be a few additional direct interfaces.
The Ethernet is a coax cable with about 2 megabit rate and a simple protocol.  
This scheme requires only one hardware interface and one software interface for each
new device instead of a hardware and software interface for every pair of 
devices.\.

\JDigitized stereo images may be viewed and processed at a stereo station.
The station is comprised of a pair of black and white monitors, 17" Conrac QQA-17C
high resolution video monitors, and a modified Old Delft III scanning
stereoscope.  The Grinnell display system provides video to the stereo station monitors.
The cursor can be manipulated by trackball
control through the PDP-11 and PDP-10 computers.\.

\JThere is a computer-controlled animation box which enables the making of
still pictures and motion picture.  The device has single frame exposure
and control of color filters, used with multiple exposures from a black
and white monitor.\.

\F0\CSAIL

\JThe system is display-oriented, with full graphics display
terminals in nearly every office.
Display editors and other system software provide unusual
user facilities.  High quality printing and graphics devices provide
publication facilities which are supported by software 
developed at the AI laboratory (PUB, TEX for
mathematical publication, and POX).
Graphics printing devices include: the XEROX Graphics Printer (XGP),
XEROX Dover Printer
(high resolution bit raster), Varian 22" printer/plotter (bit raster),
Canon laser beam printer, and Alphatype photo composition printer.
Extensive graphics packages exist for vector graphics, gray scale and
overlay, for display and for hard copy.\.

\JThe main computer is a Digital Equipment Corporation KL10 with 11 megabytes
of 36 bit words.  It has 8 drives of Ampex 3330-11 discs, 10.5x10↑9 bits.
Peripherals include a DataProducts printer, a Calcomp plotter, two 7 track
tape drives, and 4 Dectape drives.\.

\JThere are 53 displays with 32 channels of Datadisc digital raster memory,
6 III vector displays, 3 Imlac vector displays, 35 Datamedia character
displays, 15 teletype terminals, and 5 TI terminals.  The Datadisc displays
are interconnected by a digital switch which allows overlay of channels.
The Datadisc supports graphics and gray scale by overlay of channels.
The primaary graphics and gray scale display device is a Grinnell 
raster buffer display with 512x512 pixels with 36 bits per pixel.\.

\JA DEC KA10 serves as peripheral processor. A DEC PDP-11/45 with 80 kwds
of 16 bit memory and with an SPS-41 array processor acts as a real-time
processor for robotics.  A TI multiprocessor system also is used for robotics.\.

\CSCORE

\JSCORE is a DECSYSTEM-2060 running
TOPS-20.  SCORE includes 1024K words of main memory and 800 million
bytes of online storage.  SCORE supports 64 local and dial-access terminals,
and ARPAnet access.  It is jointly owned by the departments of
Computer Science,
Electrical Engineering, and Operations Research.  All students in a degree
program in Computer Science have access to SCORE.  SCORE is operated by
the Computer Science Department.\.

\CSUMEX

\JSUMEX is a dual processor DECsystem-1060 running TENEX.  It is a national
facility owned by the National Institutes of Health and managed by a
national governing board.  Applications of artificial intelligence to
problems in medicine and biology are the prime research foci of this
facility.  Students whose research work involves such applications may be
granted access to SUMEX.  The SUMEX facility supports local and dial-access terminals,
ARPAnet, and other national networks.  Additional equipment associated with SUMEX include
a DECSYSTEM-2020 and a DEC VAX-11/780.  These several systems are also on
the Computer Science Ethernet local network.\.

\CCSL

\JClose collaboration is maintained between the Electrical Engineering department and
the Computer Science department.  Research in hardware and software
computer systems is done in the Computer Systems Laboratory (CSL).  CSL is
a laboratory of the Electrical Engineering Department populated by faculty
from both Electrical Engineering and Computer Science.  CSL operates
a VAX-11/780 system for VLSI system design.  Also, CSL has a variety
of small computer systems.\.

\CNetwork

\JA BBN TIP (Honeywell DDP-516) provides connection to
the ARPANET.  Systems are interfaced to a local network, the
Ethernet, which connects SAIL, SCORE, many DEC VAX systems, DEC PDP11s,
and a network file server.

The Computer Science Department also operates several Xerox Alto personal
computers, linked together by the Ethernet communications network.  Xerox
has provided these Altos, a Dover printer, and a network file system, as
an equipment grant to the Computer Science Department.\.

\CPlans

\JStanford has developed the SUN terminal, a high resolution graphic display,
800x1000, with a Motorola 68000 processor.  Extensive software support
based on UNIX and C are being developed. 

Computer Science and CSL plan to acquire 35 SUN terminals,
10 VAX 11/750s and 15 LISP machines in 1982.  A large file server for the
entire SU network will provide economies of scale in storage.
Smaller file servers for clusters of SUN terminals are planned.  Since
cluster file servers are based on large disks, they are more economical
and provide higher performance than small disks for individual SUN machines.\.

\,
\F0\CREFERENCES      

\J[Salisbury 81]Salisbury, J.K., J.J.Craig;
"Articulated Hands: Force Control and Kinematic Issues";
Proc Joint Automatic Control Conference, Charlottesville, June 1981.\.

\J[Binford 72]T.O.Binford;
\F0"Sensor Systems for Manipulation";\F0
Conference on Remotely Manned Systems, Pasadena, September 1972.\.

\J[Soroka 80]B.I.Soroka;
"Debugging Robot Programs with a Simulator";
Proc CADCAM-8 Conference, Anaheim, CA, November l980.\.

\J[Ohwovoriole 80]Morgan Ohwovoriole;
"An Extension of Screw Theory and its Application to the Automation of
Industrial Assemblies";
Memo AIM-338 Artificial Intelligence Laboratory, Stanford University, 1980.\.

\J[Moravec 80]H.P.Moravec;
"Obstacle Avoidance and Navigation in the Real World by a Seeing Robot Rover";
AI Lab, Computer Science Dept, Stanford Univ, 1980.\.

\J[Miller 81]Miller, A. and Lowry, M.;
"General purpose VLSI chip with fault tolerant hardware
for image processing";
Proc Image Understanding Workshop, April 1981.\.

\J[Brooks 81]T.O.Binford, R.A.Brooks;
\F0"Geometric Modeling in Vision for Manufacturing";\F0
Proc Image Understanding Workshop, April 1981.\.

\J[Binford 81]T.O.Binford;
\F0"Inferring Surfaces from Images";\F0
Artificial Intelligence Journal, July 1981.\.

\J[Lowe 81]Lowe, D., T.O.Binford;
\F0"The Interpretation of Three-Dimensonal Structure from Image Curves";\F0
Proc Int Joint Conf on AI, Aug 1981.\.

\J[Binford 71]T.O.Binford;
\F0"Visual Perception by Computer";\F0
IEEE Systems Science and Cybernetics Converence, Miami, December 1971.\.

\J[Agin 72]Agin, G.J.;
"Representation and Description of Curved Objects";
Artificial Intelligence Laboratory, Stanford University, Memo
 AIM-173, CS-305, 1972.\.

\J[Nevatia 74]Nevatia, R.;
"Structured Descriptions of Complex Curved Objects for Recognition and
Visual Memory";
Artificial Intelligence Laboratory, Stanford University, Memo
⊗AIM-250, CS-464, 1974.\.

\J[Paul 72]Paul, R.;
"Modelling, Trajectory Calculation and Servoing of a Computer Controlled 
 Arm";
Artificial Intelligence Laboratory, Stanford University, Memo
*AIM-177, CS-311, November 1972.\.

\J[Shimano 78]Shimano, B.E.;
"The Kinematic Design and Force Control of Computer Controlled Manipulators";
Artificial Intelligence Laboratory, Stanford University, Memo
AIM-313, March 1978.\.

\J[Salisbury 80]Salisbury, J.K.;
"Active Stiffne Control of a Manipulator in Cartesian Coordinates";
Proc 19th IEEE Conference on Decision and Control, Albuquerque, December l980.\.