Abstract
I	The Need for Advanced Automation
II	Research Plan
	A.	The Pilot Assembly Station
		1.	Tasks
		2.	Hardware
	B.	Vision Development
	C.	Manipulator Development
	D.	Systems Support and Related Issues
III	Dissemination of Results and Relation with Industry
IV	Staff
V	Computing Facilities
VI	Budget
VII	References
VIII	Relevant Laboratory Output
	Films,THeses,Reports
		THE NEED FOR ADVANCED AUTOMATION



The benefits available  from  automation  in  manufacturing  are  now
widely  appreciated.   The  current  interest  in advanced (computer-
oriented) automation stems  from  two  sources.   The  first  is  the
decreasing   cost   of   computation   combined   with   an  improved
understanding of how to  apply  computation  to  various  aspects  of
manufacturing.  The second cause is the realization that many current
manufacturing operations, primarily  assembly  and  inspection,  will
require quite complex automation techniques.  The same considerations
apply to the automation of  tasks  such  as  laboratory  and  medical
procedures,   but  the  emphasis  has  been  (as  ours  will  be)  on
manufacturing.  kThere has been a great deal written  lately  on  the
potential  benefits of advanced automation and we will not attempt to
restate  the  points  made   so   cogently   elsewhere.    A   recent
ARPA-sponsored  study  []  in  which  we  participated  suggests  the
following major benefits from advanced automation:
	reduced prices
	increased product quality and uniformity
	customization of products
	increased job satisfaction. We would like  to  add  just  one
comment.   As we have seen with other technological developments such
as numerical-controlled tools, there will be things made  which  were
previously  impractical to make. There is widespread agreement in the
technical , industrial and government communities  on  the  potential
benefits of advanced automation.  There is much less agreement on the
level of sophistication needed in automation hardware  and  software.
We   believe  that  the  most  practical  solution  for  intermediate
quantities will be to develop general purpose assembly and inspection
devices controlled by sophisticated programs in digital computers.

We  assume  that present manipulators have a small but stable market,
and that most will have minicomputers  soon  (Sundstrand  does  now).
Will  the  incremental  cost  of  touch,  force and visual sensors be
offset by the opening of new  capabilities,  and  will  the  cost  be
offset  by  reduced  cost  of  system design for machines which allow
loose alignment tolerances?  We estimate hardware costs in a separate
paragraph,  and estimate that by 1977, the incremental cost of visual
sensors will be roughly $1100, with about $300 additional for a laser
ranging  scheme  as  we  use  now.        With  the additional memory
requirements, and the multiple of two in  marketing,  visual  sensors
will add $4k to $6k to the selling price, or about 15%-20% of current
manipulator prices.     Presumably labor costs will inflate also,  so
that  cost  comparisons  will  not  change appreciably over that time
scale unless significant  new  mechanical  designs  are  made.      A
conclusion is that sensor-equipped robots will be available at slight
additional cost.

We do not expect mechanical production costs to drop dramatically  as
electronics  costs  fall  with integration.   The most promising cost
reduction development would be high-production standard modules  such
as  robots.     As an example, consider the Stanford arm; produced in
unit  quantity,  parts,  manufacturing   and   assembly   costs   are
approximately  $10k.  The arm is much simpler than an automobile; its
potential  selling  price  in  large   quantity   is   $3k,   without
minicomputer   control.      Potentially,  a  robot  with  two  arms,
minicomputer control, and touch sensors is possible for under $10k in
quantity.   This  is  the range of price of present transfer machines
which have limited  market  because  they  have  limited  motion,  no
control,  and are thus locked out of the economy of scale.  Versatile
assembly devices have an advantage in utilization in medium scale and
job  shop  production;  their  utilization  efficiency  can be higher
because versatility allows a variety of operations  to  be  scheduled
simultaneously.     The  availability  of  computers  for  production
scheduling will allow the flexibility to be used to advantage.   With
simple  devices, transfer and assembly operations must be broken down
into small operations  with  comparable  cost  to  a  more  dexterous
system.  Robots can cut down on real estate, since several operations
can be done at one station; belts and other transfer devices  require
more  space than desirable.  A robot which can pick parts from a cart
would save considerable space.  Robots would also  have  longer  life
over  which  to amortize costs, since they could operate with greatly
changed production operations,  and  evolve  with  new  hardware  and
software  modules.    They  would have higher reliability, since they
would have monitoring, contingency action, and error recovery.   They
would be self-aligning, self-calibrating, and self-testing. There has
been  difficulty  in  meeting  the  tight  tolerances  necessary  for
operation  of  non-intelligent  devices.   It is hard to estimate the
value of systems which  work  within  current  industrial  tolerance,
rather  than  requiring radical improvements in engineering practice,
but that continuity does seem important.

Functional modularity has been emphasized in our studies;  modularity
is  clearly  possible  and  there is a powerful economic incentive to
assemble widely varying systems from one, two, or three arm  modules,
force  and  touch  modules, vision and ranging modules, with planning
and monitoring modules,  computer-aided  mechanical  design  and  job
specification  for robots on a local or network time-sharing service,
with software modularity. This approach to advanced automation  saves
much  wasted effort. The economies of scale in production of assembly
modules are matched by economies of intellectual scale:   the  design
process  is  modularized  and  easily communicable; there is a common
discipline which can be widely absorbed; developments of many sources
can  be  integrated.     The  alternative  is special design for each
operation.  We expect that at some point, design will be cheaper with
standard  robot modules than with special purpose designs, just as it
is now more economical to use mini-computer controllers than  special
purpose  controllers.     To  some extent, the dehumanized industrial
worker now fills the role of standard  module,  so  that  we  do  not
expect   discontinuous  change.    The  continuity  with  decades  of
experience with design for  human  versatility  in  assembly  can  be
relied  upon;  robots  with  some  versatility,  sensing, contingency
actions, and dexterity will provide a  transition  to  newer  schemes
which  will  evolve  more slowly. Advances which benefit special case
automation and traditional trends in automation for human  production
will  also  aid general purpose devices.  For example, new techniques
in bonding and fastening will simplify many  assembly  tasks  to  the
capability   of   robots,   while   still   too  complex  for  simple
servomechanisms. Another advantage of general purpose  techniques  is
the  ease  with which the size of the robot can be scaled up or down.
There appears to be no difficulty in making very large or very  small
assembly stations which use essentially the same programs.

The  development  of  general  purpose  assembly devices will benefit
manufacturers and a broad spectrum of industry.  There does not  seem
to  be  a  private  institution  with the competence or the desire to
undertake this program.   It is an appropriate area for  governmental
action to help meet a national need.

But given sensor-equipped robots at little extra cost, they will have
visual ability  to  sense  color,  strong  contrasts  in  brightness,
recognize  known  objects  from two dimensional outlines (top view of
flat object) or three dimensionally using depth, to a certain  extent
they  will  be  able  to  recognize  three  dimensional objects using
monocular recognition.   They will be able to visually  align  mating
parts    in    assembly    operations,    increase   reliability   by
self-calibration and flaw and disaster detection,  and  permit  large
tolerances  in parts transfer, even permitting parts to be taken from
a bin.       Several of these abilities are not yet available in  the
laboratory,   but   with  known  conditions  and  objects,  and  some
engineering (color coding) are achievable within that time span.

		VISUAL PERFORMANCE FOR APPLICATIONS

While there will be some applications for unsophisticated devices, it
seems  very  worthwhile  to develop advanced assembly devices.    Our
research is aimed in that direction.  One part of our effort will  be
vision for assembly; an assembly task will be performed using vision,
touch, and two hands.    The vision effort will  be  integrated  with
other  abilities, and while no great generality will be stressed, new
modules will be incorporated into the system, and a set of techniques
will evolve. These will depend heavily on techniques similar to those
of our recent studies in visual feedback and  hand/eye  coordination.
The  internal  model  of  the  world  will  be used to predict visual
appearance of selected areas of the scene, and a set of  verification
procedures  will  confirm or contradict the predictions.  Many of the
operations will be connected with supplying feedback to manipulation,
for alignment and insertion. These visual procedures will have models
for the parts and approximate locations.  A typical  example  is  our
current project, assembly of a water pump.  The major parts are to be
located, the screws  are  supplied  as  is  standard,  by  a  feeder.
Locating the body and cover of the pump require a gross localization,
then a fine localization of screw  holes.   Touch  provides  a  final
accuracy  and  independent coordination of the hand.   The placing of
alignment pins within screw holes can be monitored visually; although
the previous actions provide adequate accuracy.  Placing of the cover
over alignment pins can be carried out without vision.  However,  our
improved  force  sensing  will  be  useful  for  that  operation when
completed.  The insertion of alignment pins in other screw holes, and
the insertion of screws in holes will be monitored visually. Although
the assembly has been performed up to tightening of  the  screws,  by
the hand alone, this will provide a good test case for development of
vision, which will increase reliability.

To be precise in what vision software will be  implemented,  we  will
greatly  expand  our efforts in visual feedback; at present these are
adequate for polyhedra.    We will generalize these to  objects  with
with curved edges, and extend the quality of the verification modules
in  straightforward  ways  using  our  edge   detection   and   color
facilities.  These prediction techniques involve 2d shape description
techniques.   Template matching techniques  are  straightforward  and
standard;  we  expect  to make more sophisticated description modules
which have part/whole capability, of use in  planning.   Some  simple
extensions  to 3d descriptions are planned.  We will extend the power
of our region finding and color  region  analysis  programs.   Visual
input  devices  will  be  interfaced to the minicomputer and existing
modules converted.
	Vision research

Visual  tasks  connected with assembly will initially be performed as
special case problems.   As we have pointed out,  communication  with
the  machine will be necessary for job specification and instruction.
In parallel with progress on  vision  for  assembly,  to  make  these
techniques  available,  we will pursue development of visual strategy
programs. We describe separately the work on construction strategies,
and limit this discussion to visual strategy.  The domain of assembly
is particularly attractive as a subject for strategy programs,  since
the context is much richer than in more artificial environments, such
as the blocks world.  Although a much more  complex  problem  in  the
shapes  and  structural relations, in some ways, it is simpler; there
are  more  types  of  visual  information  to  use,  more  contextual
constraints,a  greater  variety  of simple verification operations to
perform.   This step is essential if intelligent assembly devices are
to be conveniently programmed for tasks. Within the division of labor
between large and small machines, this facility must  reside  in  the
central  time-sharing  facility which has large memory, file storage,
and problem-solving languages.  The system must have a model  of  the
parts  for  assembly,  and a priority from higher up of the parts and
relations to verify or perceive. It  will  be  necessary  to  have  a
representation of shape and models for usual assembly operations such
as insertion and alignment, and the ability to  make  predictions  of
expected  visual scenes and choose a sequence of operations, all this
to  enable  convenient  communication  with  the  user.   When  these
facilities  exist, the addition of new features and operations can be
incorporated smoothly, so that users can call on a growing library of
planning  modules.   The  result  of  a planning effort is a cycle of
machine planning and interaction which results in a plan network  for
the mini computer which controls individual machines.


We will continue with our work  with  description,representation  and
recognition of complex objects using depth data.  We are able to make
complete descriptions of objects such as rings and s-shaped snakelike
objects.   The  programs  can  segment  dolls,  toy  horses,  etc  in
intuitively appealing ways and describe the segments well.  The  next
efforts will be to improve and generalize the segmentation process of
describing primitive parts, to link the  descriptions  of  primitives
into  joints,  then to form object descriptions from the descriptions
of joints.    The  recognition  of  such  objects  depends  first  on
internal  modelling  of an object, then on matching perceived objects
with previously stored internal models.  Our representation is useful
in  giving  a  symbolic  description  of  objects which is useful for
visual memory. We intend also to apply large parts  of  our  programs
for   description  and  recognition  of  curved  objects  using  only
monocular intensity information.  This is a much harder problem,  but
within  our paradigm, shares much of the same code, and relies on the
same conceptual basis of local 3d models.   This  might  seem  overly
sophisticated  to  be  relevant  to  industrial robotics, since early
applications might use the simplest facilities.   Our  representation
of  shape  provides a natural way of representing shape internally in
symbolic form and in communicating  with  a  user.  There  are  other
programs  at  the  laboratory in computer graphics (OCCULT) geometric
modelling (GEOMED) which we plan to use as a basis for Computer-Aided
Mechanical Design.



We  will  continue  work on color region analysis and on a new system
which combines region  analysis  with  semantics  about  objects  and
relations.   These  programs  already  exist, but are under extensive
development to integrate more knowledge and to fit  into  a  complete
visual  system. We anticipate that they will function in more complex
environments than have been attempted in the past. There is a growing
effort  on outdoor scenes, in connection with navigation of vehicles.
While this is not directly relevant to the industrial environment, it
does  reflect the growing involvement in complex visual tasks, and is
another force for development of useful visual programs.