COMMENT ⊗   VALID 00010 PAGES
C REC  PAGE   DESCRIPTION
C00001 00001
C00002 00002	\|\\POX Title Page with big fonts
C00003 00003
C00004 00004	\|\\standard pox left margin, right margin requires minor change
C00007 00005
C00008 00006
C00018 00007	\CIntelligence
C00030 00008	\F2\CSYSTEM CONFIGURATION
C00031 00009	\F2\CACCOMPLISHMENTS
C00046 00010	\F2\CREFERENCES
C00060 ENDMK
C⊗;
\|\\;POX Title Page with big fonts
\|\\M0BDR66;\M1BDR40;\M2GRFX35;\M3METLB;\M4NONH;\M5CLAR40;
\; NGB IS NO GOOD;
\←L\+=100;\→L\;  moving the left margin for slides
\\←R\-=100;\→R\;  moving the right margin in to give the binder room
\←=+80;\→W\;
\←L\+W\→l\←R\-W\→r\;
\←r\+=10;\→>\←l\-=10;\→<\; 
\←l\→0\+=225;\→1\+=250;\→2\;
\⊂'10044;\←=35;\ε\;
\,


\F4\CEQUIPMENT SUPPORT for

\F4\CINTELLIGENT SYSTEMS

\F4\Cfor AUTOMATION



\CPROPOSAL TO 

\CHEWLETT PACKARD



\F1\Cfrom

\CSTANFORD UNIVERSITY

\CARTIFICIAL INTELLIGENCE LABORATORY


\CJuly 1983

Principal Investigator
Thomas O. Binford
Professor of Computer Science (Research)
\|\\;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\CEQUIPMENT SUPPORT FOR
\F2\CINTELLIGENT AUTOMATION SYSTEMS

\F2SUMMARY

\JThe Robotics Laboratory of the Stanford Artificial Intelligence Laboratory
proposes that Hewlett Packard Corporation support research in intelligent
systems for robotics, inspection, and manufacturing by providing an equipment
grant of ten Hewlett Packard 9836 workstations as network-oriented
individual LISP machines.\.

\F2\CEQUIPMENT SUPPORT FOR
\F2\CINTELLIGENT SYSTEMS FOR AUTOMATION


\COBJECTIVE

\JThe Robotics Laboratory of the Artificial Intelligence Laboratory
requests equipment support for a broad program of research in
intelligent systems for manufacturing, including inspection, robotics,
and precision machining.  The equipment requested is ten HP 9836
systems configured as individual LISP workstations on the Ethernet.
We also request hardware and software maintenance for these systems.

The intention of this proposal is to equip the intelligent systems
effort with sufficient individual LISP machines running PSL to provide
adequate access to LISP machines,  a uniform set of LISP machines.
Too few machines or a random collection of machines lessens their utility,
both because of the effort of learning the machine environment and making
transitions between machines, and because of the system software effort
of maintaining and developing multiple systems and attempting to retain
portability across multiple environments.\.

\F2\CINTRODUCTION

\JThe Stanford Artificial Intelligence Laboratory (SAIL)
was established by John McCarthy in 1962.  SAIL
set up a world class robotics program in 1965.
The primary focus of the Robotics Laboratory has been intelligent systems
for vision, robotics, and manufacturing.
Close collaboration has been maintained since 1965 with Prof Bernard Roth of
Mechanical Engineering, and for the last four years with Prof Robert Cannon of
the Aeronautics and Astronautics Department.  
The Robotics Laboratory is a major factor in CAMS, the Center for Automation
and Manufacturing Science at Stanford.  

Other groups in Artificial Intelligence at Stanford include the Formal Reasoning
group under Prof John McCarthy, and the Heuristic Programming Project under
Prof Edward Feigenbaum.  The province of the Robotics Laboratory is reasoning
from first principles about actions in the real world.\.

\CEDUCATIONAL PROGRAM

\JSAIL seems to lead by far in producing world leaders and experts in robotics.
A count several years ago included about 70 people trained at SAIL either as
students or visitors.  Many of these are leaders of major robotics and automation
research efforts (IBM, CMU) or distinguished researchers.  

Stanford has taught a robotics course since 1971.  Enrollment has been about
100 in recent years.\.

\CRESEARCH PROGRAM

\JIntelligent actors or robots incorporate input, planning, and output,
i.e. sensing, intelligence, and action.  Research covers this spectrum.
This research focusses on intelligence, embodied in the ACRONYM system and
its successor.  As a preliminary, sensors and action devices
will be discussed first.

Research in sensing includes computational methods in sensory technology,
application of novel sensing methods, and where necessary, device technology.
Sensor systems under investigation in collaboration with other groups at Stanford
and elsewhere include force and touch sensors for robots,
acoustic imaging sensors, and smart optical sensors.  Application of sensors
to mobile robots and to precision manufacturing are being explored.

Research in action devices includes building effectors, controlling them,
and programming them with traditional language systems.
Research is aimed at building hands and arms which can make secure grasps,
use  tools, or exert delicate forces for assembly, which enable generic
operations of parts acquisition, parts transfer, and parts mating.
This research is carried out in collaboration with other groups on campus
and nationwide, especially MIT, JPL, and the University of Utah.

Control studies include: control
of coupled, complex systems such as a three-finger hand with nine degrees
of freedom; task level control of devices in terms of position, velocity,
and forces on objects, not joint coordinates of devices; estimation of
dynamic properties of devices; adaptive control; fundamental limits of
force control, and force control 
in parts mating.   One goal is to move loads as rapidly and smoothly as 
possible  under sensor control.
Force control is embedded in a high level language, AL,
for explicit programming of assemblies.  Research is pursued 
in debugging real-time devices, user interfaces, specification of tasks,
concurrency and synchronization, error recovery and exception handling.
A large integrated robot system is implemented in AL.\.

\CAnalysis of Assembly

\JA project is underway to investigate the assembly of precision objects
by robots.  The tasks are to: 1. analyze the parts presentation, whether
to present parts by kitting or bulk feeders, according to economic criteria;
2. analyze the distribution of tasks within the assembly, whether all
operations at one station, one operation per station, or an intermediate
configuration;
3.  carry out an assembly using two coordinated manipulators;
4.  study redesign for automated assembly.\.

\CUltra Precision Machining

\JAn ongoing 
interdisciplinary project will construct a diamond turning machine accurate
in the range of .01 micrometers. This effort combines precision instrumentation,
precision mechanical actuators, artificial intelligence and computer science.
This research is intended to expand into a broad program in generic technology
for ultra precision manufacturing.  The project will bring several innovations:
1.  novel optical methods including speckle interferometry for measuring 
surface finish; 2. newly developed laminar flow hydraulic actuators and
spindle drives developed at Stanford; 3. modeling of the machine combined iwth
on-line,  real-time model-based control; 4.  rule-based high level machine operation.\.

\CIntelligence

\JThe focus of this project is on intelligence for planning and acting in
the real world, a domain with enormous scope.

An aspect of the research is study of the architecture of vision algorithms
and their implementation in integrated circuits (VLSI) to achieve
real time performance, i.e. fast enough to be useful in applications.
This study includes fault tolerant design.  Most research has been aimed
at image processing algorithms, but the intention here is to study 
implementation of high levels of perception and reasoning in high performance
VLSI.

One goal is to introduce structural learning mechanisms into ACRONYM,
to learn object models and object classes, to learn 
rules and strategies, i.e. programs, and to learn when to use what it has 
learned.
Building the knowledge base or rule bases which capture the knowledge of experts
consume great effort.
A central concern is learning from one example or few,
learning causal relations in distinction to
statistical correlations, because causal relations define decision
criteria which are dependable.  In contrast, decisions based on
statistical distributions are often unreliable because of systematic
bias. 

Representation is a central issue in defining similarity classes of
objects.  It is argued that fundamental definitions of object classes
are functional, that function = form [Binford 82].
An essential part of learning is relating function and structure.\.

\CPerception

\JPerception research aims to extend monocular cues to object interpretation
and figure/ground  discrimination.  It  is computationally  infeasible  to
attempt to match all combinations of features against models.
Instead, it is important to group image features in canonical ways. 
Segmentation and aggregation are complementary processes which discriminate
features and group them.  Some grouping is suggested by image structure,
corresponding to figure/ground discrimination.  

Other groupings arise from three dimensional interpretation of images.
This is known as inferring shape from shape, three dimensional 
shape from two dimensional shape.   Inference rules guide the estimation
of surface discontinuities (obscuration), surface tangent discontinuities
(edges), surface orientation, and surface curvature.  Inference rules for
the interpretation of shadows is also a topic for exploration.

Research has been pursued on geometric constraints in stereo vision.
These constraints are the key to building effective stereo mapping systems
which might serve the needs of the Defense Mapping System.  They include
traditional mechanisms for stereo correspondence but introduce much new
monocular interpretation.  The basic theme is to integrate all constraints,
general and specific.

Two major themes are
multi-sensor integration with collateral information and knowledge bases, and
interpretation of images by matching with generic and specific models.\.

\CGeometric Modeling and Geometric Reasoning

\JGeometric reasoning mechanisms include the rule language for
expression of rules, the rule base for transforming constraints
into algebraic constraints, symbolic manipulation routines for
performing geometric transforms and simplification to derive
symbolic prediction expressions with variables, constraint resolution
mechanisms to test satisfiability of constraint systems,
statistical constraint propagation with symbolic expressions.

Two major areas of research are problem formulation and
generation of efficient strategies.  Problem formulation will 
enable stating and solving problems such as finding efficient algorithms
for determining convex hulls and similar problems from computational
geometry.\.

\CRepresentation

\JIn geometric modeling, ACRONYM is based on generalized cylinder representation
for solids.  Complementary representations of surfaces, edges, and vertices
are being introduced, together with meta representation of geometric types.
Representations of kinematics of joints and assemblies are under investigation.
Graphics methods are explored for broader classes of generalized
cylinders.  

The representation of physics in an intelligent system is under investigation.
Physics makes an identification between physical structures and
geometric structures, e.g. representing positions as vectors,
and an identification between physical actions and geometric maps, e.g.
rotations as orthogonal transformations. 
Sensors and observation will be modeled along with kinematics and
dynamics.

Physical object classes are not defined by three-dimensional shapes, but
by function.  Mechanisms will be implemented for
representing function as structure.
One aspect of function is fabrication, representing fabrication processes.

The representation of space/time sequences in intelligent systems
makes use of the same mechanisms as the representation of space.
The representation of time assumes events and an observer
with an internal clock, i.e. a mechanism for judging before and after, with
an internal metric.   Coordinate transformations relate time frames.
Events may be continuous or discrete. 
Parallelism and synchronization are represented by constraints.

Geometry serves as the underlying representation
for the physical world, and as the underlying representation for implementation
of the intelligent system.  Relations will be represented among many levels
of geometric types which include real numbers, tuples, vectors, points,
curves, surfaces, and volumes; we plan to represent maps among these types.
Programs which manipulate geometric structures can be represented as maps.
A system which reasons about geometry should understand the structural
relationships among geometric types in order to formulate and generate programs
which solve geometric problems.  In order to reason by analogy, a system
should understand geometric programs in a deep way.\.

\CApplications

\JDevelopment of a mobile robot is underway which builds its own model of a  changing
environment, which  is intended to have  high  performance  acoustic  
imaging  and  stereo vision, and which has intelligent task planning. 

Expert systems are being developed for Image Understanding, for robotics, 
and for manufacturing.

Development is underway of a next  generation robot programming language  based
on geometric  reasoning in  ACRONYYM,  with geometric  modeling, use  of  a
computer-aided design data base, planning and problem solving,  mechanisms
for integration of  vision and  other sensing,  off-line programming  with
simulation,  and   a   hierarchy   of  language   and   computer   systems
(maxi,mini,micro).  This includes automating choice of grasping positions,
 extending 
collision  avoidance,  automating  the  use  of  sensing   in
verification of  robot actions,  and automating translation of  constraints
into positions  and forces  in parts  mating. 

Intelligent systems  are under investigation for  computer-aided  mechanical
design,  based  on   the  ACRONYM   system.  CAD  as   now  practiced   is
computer-aided drafting and archiving.  This study 
concerns creative design.
An objective is a fast  turnaround prototyping fabrication facility. 
A study will be made of synthesis of
mechanical designs using the specification of design constraints in  terms
of function and  the relation  of function  to structure  [Lowry 82],  and
methods  of   testing  constraint   satisfaction.\.
\F2\CSYSTEM CONFIGURATION

\F0The proposed system should have the following configuration:
(details to be worked out):

ten HP 9000 computer systems
  floating point hardware when available?
  4 MB memory
  100 MB disk
  terminals with graphics capability
  Ethernet link to SU network
  software
    PSL, PASCAL, C, UNIX, mail, network
  hardware and software support
\F2\CACCOMPLISHMENTS

\C\F0Intelligent systems

\JRecent research has centered on the ACRONYM system, an intelligent,
knowledge-based system for vision and planning [Brooks 79, 81].
It integrates subsystems for interaction with users, for interaction with the world,
for modeling the world, 
and subsystems for planning, interpretation, and decision making.
ACRONYM contains a substantial
core of fundamental mechanisms which are powerful and general.  Its performance
demonstrated thus far depends on domain-independent capabilities, not on
special domain-dependent tricks. 
ACRONYM incorporates geometric modeling with rule-based geometric
reasoning.

Recognition of aircraft has been demonstrated in ACRONYM.
ACRONYM has been used as a central part of a major  ARPA project and has
been distributed to about ten research groups.

Recent efforts have gone towards SUCCESSOR, a follow-on to ACRONYM.
The modeling system is being extended to include negative volumes,
multiple instances of objects, and a general name-scoping mechanism.
The representation of shape is being extended to
a very general subclass of generalized cylinders.  A powerful graphics
display system is being built.  A large set
of aircraft models have been built.  With these, some problems with the rule
set for recognition have been discovered, and the constraint resolution
system has been tested.

[Malik 82] described
representation of space/time as constraint systems and efficient methods
for resolution of constraints.
A joint effort with MIT produced preliminary results in learning
which relates object function to form [Winston 83, Lowry 82].
This work is a step toward learning of object classes and learning
of production rules, i.e. procedures,  by combining geometry in ACRONYM and work
relating function and structure at Stanford with work at MIT on learning by 
analogy and a natural language input system.

Early work led to the representation of objects by generalized cylinder
primitives [Binford 71], and showed methods of calculating generalized
cylinder representations from surface data [Agin 72, Nevatia 74].

[Agin 72] described parts of complex objects as
generalized cylinders with circular cross sections.
[Nevatia 74] built a system which recognized complex objects 
including dolls, toy horses, and tools.
The system  demonstrated "learning by showing" in three dimensions:
it built models of objects shown to it on a table, then
recognized objects when they were presented
in different orientations with different articulation and some obscuration.
They introduced new concepts for segmenting complex objects into parts,
for building part/whole models
of complex objects as structures of generalized cylinders.  For
efficient matching to models, they used coarse stick figure descriptions
to select subclasses of similar objects for detailed matching.\.

\J[Baumgart 74] built GEOMED,
a geometric modeling system based on a polyhedral representation, using
Euler's theorem and  a "winged edge"
representation.  GEOMED was named for its powerful, interactive 
geometric editor.\.

Recent theoretical results outline a general theory of interpretation of 
line drawings which promises to include arbitrary numbers of curved surfaces
with surface markings and shadows, paper (non-solids), cracks, and wires
[Binford 81]. These results suggest general rules for inferring whether
surfaces are planar, or if not planar, for inferring their curvature.
Results also show effective means for using shadows as human interpreters do
[Lowe 81]. 

[Lowe 82, Binford 82] describe extensions of this approach
to general methods for grouping of image features as ribbons and
grouping operations for continuous curves.  These methods address the
figure/ground problem; they are central to subsequent recognition.
reconstruction of three-dimensional shapes based on shadows;
They also present preliminary results in finding clusters
of dots, edges, and texture elements.  

[Miller 82b] describe architecture studies aimed at development of
an array of 128x128 bit-serial general purpose processors in wafer scale
integration.  There is a large probability that some of the processors
will be faulty.
Such a large circuit requires fault tolerant design.  They demonstrate
an algorithm for connecting the array around defective processors, an algorithm
effective for up to 10% defective processors.  They describe
software experiments used to explore issues of architecture.
Previously, [Miller 81] designed, implemented, and tested one working element of
the array. SAIL has surveyed hardware implementation of 
vision algorithms. \.


\CStereo Vision

\JThe group made a comprehensive survey of stereo systems and available techniques.
The group has a major effort underway to implement and integrate all available
constraints and knowledge, general and special, monocular and stereo,
within a comprehensive, rule-based stereo interpretation system.
Most of these constraints, especially monocular cues, have not been included 
in any previous stereo system; all others use only a few, simple conditions.
The constraints of [Arnold 80] have been included in a system with a dynamic
programming search process [Arnold 83].  Geometric constraints include:\.
\jangles of corresponding views of a vector;\.
\jprojections of corresponding views of a surface;\.
\ja necessary condition for obscuration;\.
\jviews of opaque surfaces.\.

\JThe system of [Baker 81] has been extended to include the difficult case in 
which images do not correspond along scan lines.  It has been modified to
use edge data from [Marimont 82].
In order to calculate stereo geometry in a
sufficiently general way, a new system for stereo registration has been developed.
A new search method for stereo registration was developed [Clarkson 81].
Currently, the group is designing generic models of buildings to serve
in a rule-based system which incorporates monocular and stereo inference
rules based on [Binford 81, Lowe 81].  Preliminary results have been obtained
in using cube corners, horizontal surfaces, and horizontal edges in
stereo correspondence, going beyond the results of [Liebes 81].

A kinematic simulator was made for robots, based on ACRONYM [Soroka 80].
A dynamic simulator for mechanical system has been built.
A rule  base  for
automated grasping in ACROMYM was made in  1979. 

Extensive experimentation has been carried out 
with touch sensing [Binford 72].  The Scheinman force  sensor was developed;
it has  been widely  distributed.  Force sensing fingers have been developed.

The group developed the AL system for programming robots, developed an
integrated experimental robotics system based on AL,
then made  a portable version.  
There  are  three  other  implementations  based  on  AL,  at
Karlsruhe, Tokyo Univ, and Univ of  Grenoble.\.

\JSAIL designed  and  built  the  Scheinman  Stanford  arm  in  1970  and  the
Stanford-JPL three finger hand [Salisbury 81].  Both have had  significant
impact outside Stanford;  several laboratories  are buying  copies of  the
Stanford-JPL hand.   The Salisbury  hand and  Okada's hand  at ETL,  Japan
represent the  state of  the art.
A hydraulic arm was built in 1968; Scheinman built the 
Stanford in 1970.  
Paul implemented WAVE in 1971, the first programming system 
to provide several modern mechanisms.
The VAL programming system of Unimation
was based on WAVE; Bruce Shimano, formerly of this laboratory, has led
VAL development since its inception.

The arm and WAVE were combined in the first 
computer integrated assembly, the assembly of an automobile  water
pump from 10 component  parts [Paul 73, Bolles 73], 
using force and touch sensing extensively along with vision, 
a power  tool, and several fixtures.

[Salisbury 81] made  a sophisticated  theory  for  coupled
control of the three finger hand using tension sensors.
The hand has three fingers,  each with three degrees  of freedom.  The control
commands the position, velocity and  internal force on an  object to be grasped
by the hand, rather than command individual motors for nine joints.  The same
analysis is  relevant  for  two  arms cooperating  or  other  closed  loop
kinematic chains. 
In the past SAIL implemented compliance and force control based on
a general analysis [Salisbury 80]. 
[Khatib 83] has  implemented a Cartesian  force control system, with real-time
collision avoidence.  The  Cartesian system  facilitates real-time  collision
avoidance. 
The Cartesian  force  control
system has  no known  rivals.  
Collision avoidance has been investigated by [Widdoes 74].

SAIL has studied
navigation of mobile robots [Schmidt 72, Moravec 80].  A mobile robot
demonstrated navigation around obstacles in realistic interior and external 
environments, using stereo vision [Moravec 80].\.
\F2\CREFERENCES

\F0\J[Agin 72] Agin, G.J.;
\F1"Representation and Description of Curved Objects"\F0
AI Lab, Stanford University, Memo AIM-173, CS-305, 1972.\.

\J[Arnold 77] Arnold,R.D.; \F1"Spatial Understanding"\F0
Proc Image Understanding Workshop, April 1977.\.

\J[Arnold 78] Arnold,R.D.; \F1"Local Context in Matching Edges for Stereo
Vision"\F0; Proc Image Understanding Workshop, Boston, May 1978.\.

\J[Arnold 80] Arnold, R.D., Binford, T.O.;
\F1"Geometric Constraints in Stereo Vision"\F0
Proc SPIE Meeting, San Diego, Cal, July 1980.\.

\J[Arnold 83] R.D.Arnold;
\F1"Geometric Constraints and Solution Mechanisms in Stereo Correspondence";\F0
Stanford AI Memo AIM-  1983.\.

\J[Bajcsy 72] Bajcsy, R.;
\F1"Computer Identification of Textured Visual Scenes"\F0
Artificial Intelligence Laboratory, Stanford University, Memo
AIM-180, October 1972.\.

\J[Baker 80] Baker, H.;
\F1"Edge-Based Stereo Correlations";\F0
Proc Image Understanding Workshop, Baltimore, April l980.\.

\J[Baker 81] Baker, H. and Binford, T.O.;
\F1"Depth from Edge and Intensity Based Stereo";\F0
Proc Int Joint Conf on AI, Aug 1981.\.

\J[Baker 82] Baker, H.;    
\F1"A system for automated stereo mapping";\F0
Proc. IU Workshop Image Understanding  September 1982.\.

\J[Baumgart 74]  Bruce Baumgart;
\F1"Geometric Modeling for Computer Vision";\F0
AI Lab, Stanford University, Memo AIM-249, 1974.\.

\J[Binford 71]  T.O.Binford;
\F1"Computer Visual Perception";\F0
Invited Paper, Systems, Man and Cybernetics Conf, Miami, Fla, 1971.\.

\J[Binford 74]  T. O. Binford et al;
\F1"Exploratory Study of Computer Integrated Assembly Systems";\F0
AI Lab, Stanford University, 
First Report, 1974.\.

\J[Binford 76] T.O.Binford et al;
\F1"Exploratory study of Computer Integrated Assembly Systems";\F0
AI Lab, Stanford University, Third Report, 1976.\.

\J[Binford 81] Binford, T. O.;
\F1"Inferring Surfaces from Images";\F0
Artificial Intelligence Journal, August, 1981.\.

\J[Binford 82] T.O.Binford;
\F1"Segregation and Aggregation: the Figure/Ground problem";\F0
Proceedings of the Rank Prize Fund Conf on Machine and Biological Processing
of Images.\.

\J[Bolles 73]  Robert Bolles, Richard Paul;
\F1"The use of Sensory Feedback in a Programmable Assembly Systems";\F0
AI Lab, Stanford University,
Memo AIM-220, 1973.\.

\J[Bolles 76] Robert C. Bolles,
\F1"Verification Vision Within a Programmable Assembly System";\F0
AI Lab, Stanford University,
Memo AIM-295, 1976.\.

\J[Brooks 79] R. Brooks, R. Greiner, T.O.Binford;
\F1"ACRONYM: A Model-Based Vision System";\F0
Proc Int Jt Conf on AI, Tokyo, Aug 1979.\.

\J[Brooks 79b] Brooks, R.A.;
\F1"Goal-Directed Edge Linking and Ribbon Finding";\F0
Proc Image Understanding Workshop, Palo Alto, Calif, Apr 1979.\.

\J[Brooks 81] R. Brooks;
\F1"Symbolic Reasoning among 3-D Models and 2-d Images"\F0;
Artificial Intelligence Journal, August 1981.\.

\J[Clarkson 81] Clarkson, K.L.;
\F1"A Procedure for Camera Calibration";\F0
Proc Image Understanding Workshop, April 1981.\.

\J[Finkel 74] R. Finkel, R.H.Taylor, R.C.Bolles, R.Paul, and J.A.Feldman;
\F1"AL, A Programming System for Automation";\F0 Memo AIM-243;
AI Lab, Stanford University, 1974.\.

\J[Gennery 77] Gennery, D.B.;
\F1"A Stereo Vision System for An Autonomous Vehicle";\F0
Proc 5th International Joint Conf on Artificial Intelligence,
MIT, Boston, Aug 1977.\.

\J[Gennery 79] Gennery, D.B.;
\F1"Object Detection and Measurement using Stereo Vision";\F0
Proc Int Joint Conf on AI, Aug 1979.\.

\J[Gennery 80] Gennery, D.B.;
\F1"Modeling the Environment of an Exploring Vehicle by means of Stereo Vision";\F0
AI Lab, Computer Science Dept, Stanford Univ, 1980.\.

\J[Goldman 83] Ron Goldman;
\F1"Design of an Interactive Manipulator Programming Environment";\F0
Memo AIM-350, Stanford University, 1982.\.

\J[Grossman 76] David D. Grossman;
\F1"Monte Carlo Simulation of Tolerancing in Discrete Parts Manufacturing
 and Assembly";\F0
Memo AIM-280,
AI Lab, Stanford University, 1976.\.

\J[Hannah 74] Hannah, M.J.;
\F1"Computer Matching of Areas in Stereo Images"\F0
Artificial Intelligence Laboratory, Stanford University, Memo
AIM-239 1974.\.

\J[Hueckel 69] M. Hueckel;
\F1"An Operator which locates edges in digitized pictures";\F0
J.Assoc Computing Mach, 1969.\.

\J[Hueckel 73] Hueckel, M.;
\F1"A Local Visual Operator which Recognizes Edges and Lines";\F0
J.Assoc Computing Mach, 20,634 (1973).\.

\J[Horn 72] B.Horn;
\F1"The Binford-Horn Edge Finder";\F0
MIT AI Memo 172, 1972.\.

\J[Khatib 83] O. Khatib;
\F1Dynamic Control of Manipulators in Operational Space"\F0;
submitted to IFTOMM Congress, 1983.\.

\J[Liebes 81] Liebes, Jr., S.;
\F1"Geometric Constraints for Interpreting Images of
Common Structural Elements: Orthogonal Trihedral Vertices";\F0
Proc Image Understanding Workshop, April 1981.\.

\J[Lowe 81] Lowe, D. and Binford, T.O.;
\F1"The Interpretation of Three-Dimensonal Structure from Image Curves";\F0
Proc Int Joint Conf on AI, Aug 1981.\.

\J[Lowe 82] D. Lowe, T. Binford;
\F1"Segmentation and Aggregation";\F0
Proc ARPA IU Workshop, 1982.\.

\J[Lowry 82] M. Lowry;
\F1"Relating Structure and Function";\F0
Proc ARPA IU Workshop, 1982.\.

\J[Lowry 82]
Michael Lowry and Allan Miller;
\F1"Analysis of low-level computer vision
algorithms for implementation on a VLSI processor array";\F0
Proc. IU Workshop Image Understanding  September 1982.\.

\J[Miller 81] Miller, A. and Lowry, M.;
\F1"General purpose VLSI chip with fault tolerant hardware
for image processing";\F0
Proc Image Understanding Workshop, April 1981.\.

\J[Miyamoto 75]  E.Miyamoto, T.O.Binford;
\F1"Display Generated by a Generalized Cone Representation";\F0
Conf on Computer Graphics and Image Processing, Anaheim,
Cal, 1975.\.

\J[Moravec 77] Moravec, H.P.;
\F1"Towards Automatic Visual Obstacle Avoidance"\F0
Proc 5th International Joint Conf on Artificial Intelligence,
MIT, Boston, Aug 1977.\.

\J[Moravec 79] Moravec, H.P.;
\F1"Visual Mapping by a Robot Rover",\F0
Proc Int Joint Conf on AI, Aug 1979.\.

\J[Moravec 80] Moravec, H.P.;
\F1"Obstacle Avoidance and Navigation in the Real World by a Seeing Robot Rover";\F0
AI Lab, Computer Science Dept, Stanford Univ, 1980.\.

\J[Nevatia 74] Nevatia, R.;
\F1"Structured Descriptions of Complex Curved Objects for Recognition and
Visual Memory"\F0
AI Lab, Stanford University, Memo AIM-250, CS-464, 1974.\.

\J[Nevatia 76] Nevatia, R.K.;
\F1"Depth Measurement by Motion Stereo"\F0; Comp. Graph. and Image Proc.,
vol.5 , 1976, pp. 203-214.\.

\J[Ohwovoriole 80] Ohwovoriole,Morgan;\F1"An Extension of Screw  Theory and  
its Application  to the  Automation of  of Industrial Assemblies"\F0;
Ph. D. Thesis, Dept Mech Eng, Stanford University, 1980.\.

\J[Paul 72] Richard Paul;
\F1"Modelling, Trajectory Calculation and Servoing of a Computer Controlled 
 Arm";\F0 Memo AIM-177,
AI Lab, Stanford University, 1972.\.

\J[Paul 73] Richard Paul, Karl Pingle, and Bob Bolles; \F1"Automated Pump
   Assembly";\F0 16mm color film,
AI Lab, Stanford University, 1973.\.

\J[Paul 76] R.Paul and B.Shimano;
\F1"Compliance and Control";\F0
Joint Automatic Controls Conf, Purdue Univ, July 1976.\.

\J[Pieper 68] Donald L. Pieper;
\F1"The Kinematics of Manipulators under Computer Control";\F0
Memo AIM-72,
AI Lab, Stanford University, 1968.\.

\J[Salisbury 80] Salisbury, J.K.;
\F1"Active Stiffness Control of a Manipulator in Cartesian Coordinates";\F0
Proc 19th IEEE Conference on Decision and Control, Albuquerque, l980.\.

\J[Salisbury and Craig] Salisbury, J.K. Craig, J.;
\F1%"Articulated Hands: Force Control and Kinematic Issues"\F0;
Proc JACC, Charlottesville, Va, 1981.\.

\J[Soroka 80] Soroka, B.I.;
\F1"Debugging Robot Programs with a Simulator";\F0
Proc CADCAM-8 Conference, Anaheim, CA,  l980.\.

\J[Shimano 78] Bruce E. Shimano;
\F1"The Kinematic Design and Force Control of Computer Controlled
Manipulators";\F0
Stanford AI Lab Memo AIM 313,  1978.\.

\J[Shirai 71] Y. Shirai and M. Suwa;
\F1"Recognition of Polyhedrons with a range finder";\F0
Proc Int Jt Conf on AI, London, 1971.\.

\J[Taylor 76] Russell Taylor;
\F1"Synthesis of Manipulator Control Programs From Task-Level Specifications";\F0
Memo AIM-282, 
AI Lab, Stanford University, 1976.\.

\J[Widdoes 74]  L.C.Widdoes;
\F1"A Heuristic Collision Avoider for the Stanford Arm";\F0
Internal Note,
AI Lab, Stanford University, 1974.\.

\J[Winston 83] P.Winston, T.Binford, B.Katz, M.Lowry
\F1"Analogy in Visual Learning";
Stanford AI Memo , MIT-AI memo .., 1983.\.