Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter.
Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.
OCR for page 279
TEL==RATION, TEIEEFF5ENCE, AND I~O=llCS:
RESEARCH NEEDS FOR SPACE
Thamas B. Sheridan
l~WOllON
The Need and the Dile
One of the dramatic dhall~es ~ by E pace is versatile inspection
ark] manipulation rankly Spent by man. Same people within art
outside NA5A wed like to automate everything but c~ot--beca~e so
many tasks are unpredictable arm therefore not doable by
~3ial-pur~se or prepr=3ra=able marines, or are one~f-a-ki~ such
that Medical automatic deviC--c to do them awe too costly ~ weight
and dollars. So human perception, plarming are, contra are required.
But to place man Physically there is consJcra bed by hazard and high
cost of life support. Remote inspection and manipulation by man, on
the other hand, poses s ~ icus problems of her getting sufficient
sensory information an] controlling with sufficient dexterity.
Artificial sensing, intelligence and control can help.
Unfortunately we have hardly begun to understand how to integrate human
and artificial brands of sensing, cognition and actuation. One t
is clear, however: to cast the problem in terms of humans versus
robots is simplistic, unprc~uctive and self-defeating. We should be
concerned with how they can cooperate.
Definitions
Telecperation is extension of a person's sensing and manipulating
capability to a location remote fr all him. A t~leoperator includes at
the munimNm artificial sensors, arms and hands, a vehicle for carrying
these, and communication channels to and from the human cgerator.
Telepresence is the ideal of sensing sufficient information about
the teleoperator and task, and communicating this to the human cFerator
in a sufficiently natural way that she feels herself to be physically
present at the remote site. A more restrictive definition requires, in
addition, that the teleoperator's dexterity match that of the
bare-handed human cooperator.
279
OCR for page 280
280
Robotics is the science and art of perfonnirP;, by mans of an
autistic apparatus or device, functions ordinarily ascribed to human
beings, or operating with what appears; to be almost human intelligence
(adapted from W~bster's 3rd Al. Dictionary).
Telerobotics is a form of teleoperation In Rich a human c~pera~r
acts as a su~risor, ~runicating to a ~u~r information abut
task goals, constraints, plans, contingencies, assumptions, suggestions
and orders;, getting back information about ac~arrplis~nts,
difficulties, concerns, and, as requested, raw sensory data- while the
subordinate telec~perator executer the tack based on information
received from the human operator plus its awn artificial sensing and
intelligence. A~nying the human supervisor is a ~u~' which
can c~mn~nim~te, mt~te, assess, predict, ark arise ~n
human-fri~y teens; at the site of the ~lerobot is a ~u~r which
can c~nicate with the human-~-eractive computer and effect control
using the artificial sensors and effecters In the most efficient way.
One human~7utPr Remand station can sunrise many t:elerobots.
Supervisory control in the present context is mostly synonymous with
teler~tics, referring to the analogy of a human supervisor directing
and x~nitorir~ He activities of a human suborn. Su~risory
control does not necessitate that He subordinate person or machine be
rate.
~;`rly History
Prior to 1945 there were crude tPle~rators for earth moving,
constnlction and relate task;. About that time the first moment
mastPr-slave teleoperators were develop by Goertz at Argonne National
Labs. These were mechanical pantograph mechanisms by which radioactive
materials in a "hot cell" card be manipulated by an operator outside
the cell. Electrical and hydraulic servomechanisms soon replaced the
Direct Micas tape and cable linkages (Goertz, 1954), and closed
circuit television was introduced, so that now the operator cculd be an
arbitrary distance away. Soon tPlemanipulators were being attached to
sub marines by the Navy and used commercially by offshore oil extraction
and m~hle-laying firms to replace human divers, especially as
cremations got deeper. By the mid 50s technological developments in
"telepresence" (they didn't call it that at the time) were being
demonstrated (Mouths, 1964; Johnsen and Corliss, 1967; Heer, 1973~.
Among these were: force reflection simultaneously in all six degrees
of freedom; hands with multi-jo~nt^~ fingers; coordinated two-arm
telecperators; and head-mounted displays which drove the remote camera
position and thereby produced remarkable visual telepresence.
By 1965 experiments in academic research laboratories had already
revealed the problems of teleman~pu~ation and vehicle control through
time delay (Ferrell, 1965), and the early lunar teleoperator Surveyor
demonstrated the problems vividly On an actual space mission. Touch
sensing and display research was already underway (Strickler, 1966)
though there was little Interest in teletoubh at that time. Soon
thereafter supervisory control was shawn to offer a way around the time
e
OCR for page 281
281
delay problem, and also to have advantages even without time delay in
the communication channel, where, in order to avoid collision or
dropping grasped objects, quicker teleoperator reaction time was needed
than the distant human operator could provide (Ferrell and Sheridan,
1967).
though the NASA nuclear rocket project mounted
a major effort in
teJeoperator development in the 3960s, after that program was canoelled
and throughout the 1970s there was little support for space
teleoperation or telerobotics. By 1970, however, industrial robotics
was coming into full swing, for Unimation, GE and a handful of other
American, Japanese and Scandinavian manufacturers had begun using
relatively simple assembly-line robots, mostly for spot welding and
pa mt spraying. By 1980 industrial robots had become graced by wrist
force sensing and primitive computer vision, and puch-button "teach
pendant" control boxes were being used for relatively simple
programming from the shop floor.
Overview of Current Status
To outward appearances sex-degree-of-freedom, force-reflecting,
serial-link electrical or hydraulic mast~r-slave manipulators have
changed little in forty years. There are a few new and promising
mechanical configurations of arms and mwiti-fingered hands in
laboratories, but as yet they are unproven On practical application.
Video, driven by a demanding marketplace, is now of high quality and
miniaturized, and digitization and simple recognition processing of
video images is fast and inexpensive. We have a variety of touch
(surface contact and pressure array) sensors available in the
laboratory, but as yet little under standing of how to use these
sensors. On Operation depth perception remains a serious problem,
but there is promising research on several fronts. We still have not
achieved fine, dexterous t~lemanipulation with high fidelity feedback
as implied by the term presence".
As yet there is no satisfactory control theory of manipulation as an
integrated sensory-in ator control activity, but new theories have been
develcged for manipulation task-analysis from an AI perspective, for
kinematic-4ynamic control of complex linkages, and for
force-displacement hard-en vironmant impedance. We still think of
controlling manipulator arms and the vehicles which carry them as
separate activities; we haven't learned to combine the two (though
infants do it with eased. We have demonstrated simple
human-supervised, comput~r-aided Operation in a number of ways, but
our understanding of human-comput~r cooperation is very primitive,
hardly commensurate with the label "~1erobot" we employ with such
abandon.
OCR for page 282
282
SPECIFIC AREAS IN WHICH NEW RESEARCH IS NEEDED
Research needs are discussed in four categories: (1)
(2) teleactuatinq, (3) ccm~uter-aidina in sucervi50rY
t=1~,
control, and (4)
meta-analysis of human/co mputPr/telecperator/~.~k interaction. Some
recent an] current research is cited.
Tel~nsing
My colleague, Dr. Stork, who is an MD an] more sense-able than I, will
deal more extensively with this category, particularly with vision, the
most Important human sense, and with the needy= and possibilities in
virtual displays and controls, depth perception, and other significant
needs in teleoperator research.
I would like to comment about resolved force, touch, kinesthesis,
proprioception, and proximity- five critical teleoperator sensing needs
which must be recognized as being different Fran one another. These
five, together with vision, are essential to achieve the ideal of
"telepresense". For each it is important to urxierstand how the human
normally functions, and then to urxierstand how the appropriate signals
can be measured by artificial transducers and then displayed to the
human Operator arx3/or use by artificial tntelliger~ce ~ a way helpful
to ache human operator.
Resolved force sensing is hat the humanbody's joint, muscle and
tendon receptors do to determine be net force and torque acting on the
hand, i.e., the vector resultant of all the component forces and
torques operating on the environment. In force reflecting master-slave
systems this is measured either by: (1) strain gage bridges in the
wrist (so-called wrist-force sensors); (2) position sensors in both
master and slave, which, when compared, indicate the relative
deflection in six DOF (which in the static case corresponds to force)
(3) electrical motor current or hydraulic actuator pressure
differentials. Display of fP=~hack to the operator can be
straightforward in principal; in force-reflecting mast~r-slave systems
the measured force signals drive actors on the master arm which push
back on the hand of the operator with the same forces and torques with
which the slave pushes on the environment. This might work perfectly
In an ideal world where such slave-back-tc-~aster force serving is
perfect, and the master and slave arms impose no mass, compliance,
viscosity or static friction characteristics of the Or own. Unhappily,
not only does reality not conform to this dream; it can also be said
that we hardly understand what are the deleterious effects of these
mechanical properties in masking the sensory information that is sought
by the operator In E~f~mi~ telemanipulation, or how to minimize
these effects. At 1~t, Shard to ~u~r coordinate transformation,
it has been shown ~t master arKi slave need not have the same
kinematics (Corker ark Bejczy, 1985~. Force reflection can also be
applied to a rate-con~crol jc~rstic~k (Lynch, 1972) but it is less clear
what the advantages are.
;
OCR for page 283
283
Tough is the hem use sloppily to refer to various forms of force
sensing, but more precisely to refer to differential pressure sense of
the skin, i.e., tile ability of the Skin to detect force patterns, with
re ~ to displac ~ nt both tangential and normal to the skin surface,
and to time. The skin is a poor sensor of absolute magnitude of force
normal to the surface and it adapts quickly. There are now a few
instruments for artificial teletouch; most of these have much coarser
spatial resolution than the skin, though a few of the newer ones
utilizing optic have the potential for high resolution (Harmon, 1982;
Schneiter and Sheridan, 1984~. A major research problem for teletouch
is how artificially sensed pressure patterns should be displayed to the
human operator. One would like to display such information to the skin
on the same hand that is operating the joystick or master arm which
guides the remote manipulator. m is has not been achieved
successfully, and most success has been with displaying remote tactile
information to the eyes using a compuber-graphic display, or to skin at
some other location. .
Kinesthesis and proprioception are terms often used together, at
least ~ part because the same receptors On the human body's muscles
and tendons mediate both. Kinesthesis literally is the sense of motion
and proprioception is awareness of where in space one's limbs are.
Telekinesthesis and t=1eproprioception are particularly critical
because, as telemanipu~ation experience has shown, it is very easy for
the operator to lose track of the relative position and orientation of
the remote arms and hands and how fast they are moving in what
dir ~ ion. This is particularly agg~-ava1:ed by his having to ~ erve
the remote manipulation through video without peripheral vision or very
good depth perception, or by not having master-slave position
correspondence, i.e., when a joystick is used. Potential remedies
are: m ~ tiple view ; wide field of view from a vantage point which
includes the arm base; and computer-generate] images of various kinds
(the latter will be din further below). Providing better sense
of depth is critical to t~lemanipulation in space.
Proximity sensing IS not someth mg humans normally do except by
vision, but cats do it by whiskers or olfaction (smell), and bats and
blind persons do it by sound cues or vibrations felt on the face.
Sonar, of course, will not work in space. Electrc magnetic and optimal
systems can be used for measuring proximity (close- ~ rang Meg) to avoid
obstacles or decide when to slow down in approaching an abject to be
man~pula~ (Bejczy et al.980~. Such auxiliary information can be
displayed to the eyes bar means of a ~u~r graphic display, or, if
the eyes are considered overloaded, by scars patterns, especially
ter~generated Seth. We need ~ unders~cand how best ~ use such
information In space.
TELEACTUNIING
It was stated In the previous section that we know relatively little
about certain types of rate sensing, i.e. , both artificial sensing
arm display to the human operator controlling We tele~ator (this in
spite of gnawing a gnat d-~1 abaft human sensing per se). Rate
OCR for page 284
284
ac~tion (in which ~ we include control In the conventions sees=)
poses an even larger prdblen, since it combines motor ac~tion with
sensor and decision-raking, and it can be said we Maw even less about
this, except for the practiced knowledge we have from Operating the
kinds of t=le~rators that have been art for a number of year;,
By ~ clear hot-laboratories and for undersea oil operations.
Again, Ants are offer In a ~ of specific categoric= where
some research is ongoing but Ah more nets to be done. The conch
prcble~ns in this category, where Aural-=' interaction per se is not the
pr~cip~1 issue, apply to both dint ant supervisory control.
Multi - =ree~f-fre~c~m ~~ffectors seem a most obvious need, as
evidenced by our ~ human harts, but the sad fact is Cat these have
not been develops beyond a for laboratory prototypes. Dial
manipulators ted to have simple parallel-jaw grippers, art a few have
claws, magnetic or at suction gripping mechanisms, or special pur~se
atta~nt device= for welding, paint spraying or other sE>ecial-purpose
tools. Though parallel-jaw griper s ~ ; the most obvious function
for a one DOF end-effecLor, it is not yet clear what a second DOF right
be for, or a third, etc. Mh~ti-fingered devices such as those by
Salisbury (1986) or Jacobson (1987) will help us answer these
questions. At the moment fear of losing objects in space seems to
militate against general purpose grippers; that could change in the
future. Modern computer-graphic workstations begin to offer the hope
of studying problems like these by computer simulation without having
to build expensive hardware for every configuration and geometric
relationship to be tested.
Two-arm interaction is a necessity for much human manipulation (it
has become standard for nuclear hot-lab manipulators), but we rarely
see it ~ industrial or undersea teleoperators. Part of this problem
is to get the most out a given number of degrees-of-freed=~. For
example, instead of having a single s =-axis arm operating on one body
relative to a second body for base), one ~ got acccmplish the same by
having a three DOF "grabber arm" position the body so that a second,
say, three DOF arm can work an coordinated fashion to perform same
assembly task. Industrial robot experience shows that two three DOF
arms are likely to be simpler and cheaper that one s~x-DOF arm. This
has not been implemented in space applications; the problem needs
regear=.
Redundant DOF Hand-arm-vehicle coordination is a serious problem,
and actually a need for any kinematic linkage of more than Six DOF
which must be controlled in a coordinated way. This is largely an
unsolved theoretic problem, at le=t In part because the number of
configurations which satisfy given end-point position/orientation
constraints Is infinite. One tries to select five amoral scheme
solutions to minimize energy or time or to avoid ~ Stain a ~ olute
positions of the joints, or to prevent singularities, etc., but the
mathematics is formid~hie. One arm of three and one of four DOF make
for such redundancy, but perhaps even more important, so does a vehicle
thrusting in six DOF with an attached arm of even one DOF. We humans
coordinate movements of our own legs, arms, and bodies (many redundant
OCR for page 285
285
DOF) without difficulty, but just how we do it is still a relatively
wel1-k_pt secret of nature.
`~-i-person cooperative control is one way to control a complex
mulci-30F teleoperator--where each of several Operators is responsible
for maneuvering a single arm or vehicle ~ relation to others. Is this
best or is it better to have a single operator control all DOF of both
vehicle and arm? lie really don't Bylaw. Results fray simple tracking
experiments suggest Off control of multiple independent tasks is very
cliff cult for one person. Men the Ins of freedom of a task are
closely coupled arbor ~st be coordinated to achieve the task
Objectives, ~t can be relatively refly provided proper control Beans
are pr~rided--but up to how many DOF? It is surprising how little
research is available in this area.
Adiust~hle impedance of issuer arc/or slave is a P~miSi~ way of
kid a m~ster-slave teleoperator more ~ Battle than 1~- the
compliance-viscosity- Instance parameters remained fixed (Raju, 1986).
A carpenter may carry and use within one bask several different
hammers, and a golfer many clubs, because each provides an impedance
characteristic appropriate for particular tasks which are expected.
Carrying many teleoperators into space may be avoided by making the
impedance between slave and task and/or between human and master be
adjustable. We have hardly begun to understand this problem, and have
much to learn.
Interchangeable end-effector tools is another way to accomplish
versatility and of course is Precisely what carpenters, surgeons or
other craftsmen use. Future space Operators may have a great
variety of special tools for both modifying and measuring the
environment. It is not clear how to make the trade between special and
general purpose end-effectors.
Task-resolved manipulation means performing standard or
preprogrammed cgerations (e.g., cleaning, inspecting, indexing a tool)
relative to the surface of an environmental object (Yoerger, 1986~.
This means sensing that surface in the process of manipulating and
cont~nu~ly performing coordinate transformations to update the axes
with respect to which the operations are being done. This is an
extension of point resolution" ability to cc~mr~ the finger to
move ~ a desired trajectory blithe having to worry about hear to mere
all the joints in between.
Force-fe~back with time delay has been show both theoretical By and
experimentally not to work if the force is fed back continuously to the
same hand as is cgerating the control, for the delayed feedback simply
forms an inst~hili~ on the process which the operator might otherwise
avoid by a mave-and-wait strategy or by supervisory control (Ferre1l,
1966~. Yet it sums that forces sully encountered or greater than a
preset magnitude might be fed back to that ha m for a brief period,
provided the forward gain were reduced or cut off during that same
brief period, and the master then repositioned to where it was at the
start of the event with no forcerfeedback.
OCR for page 286
286
~ut~aidir~ ~ S~risory Control
C~uters may be use for relatively "la~r-level" Stations In
many of the telesensing/display and Actuation modes describe
abc~v~. There are a .-nm~er of other tel~peration ~ problems in
Rich the b^=nan~uter interaction is He i~r~cant part. These
include c~uter simulation, ~uter-bas~ pla~/decision-aidir~,
arm c~put~r-aided ~~mmni~ation~control ~ various mixes. All
of these are pare of supervisory control by a huran operator of a
telenil30t.
Off-~ne, real-time, human~rable ("flyable") simulation of
ieleoperation for research, er~n~'ingortrainir~has barely Pronto
be viable e This Is because of the c~nplexity of simulating and
displaying the vehicle plus the arm and hard plus the manipulated
object plus the er~viromnent, having all degrees of freedman cooperate,
with removal of hidden lines, arm so on. Even namin~ly high~ality
cc ~ use r graphics machines have trouble with generation of such complex
displays in real time. We can come close today, but since computer
power is the one thing that is bound to improve dramatically over the
course of the coming few years, we ~ ght pay attention to the many
possibilities for using computers as a substitute for building
expensive hardware to perform man-machine experiments and evaluate new
design configurations. me are serious problems to simulate the full
dynamics of Anti DOF arms and harts. There He problems to be solved
to man simulated tel~rators grasp and manipulate si~at~
objects. m ere are many problems to get high ~ lity pictures (in
terms of resolution, frame rate, gr~y-scale, color, etc.) Telepresence
is an ides in simulators just as it is in actuality. In fact, to
enable the human operator to feel he is ·ltherell when lltheretl exists
nowhere other than in the computer poses a particularly interesting
challenge.
On-Line ~n-sibu Banning simulators might be used "in the heat of
batt1et' to try out maneuvers just before they are committed for real
action (and real expenditure of precious rescurce5 in space). In this
case commands WoNid be sent to the co~puter-based model of the vehicle
and/or manipulator and th ~ would be observed by the operator
prospectively, i.e., before further commands are given (as compared to
the retrospective state estimation case to be described below).
Ccc=ands (supervisory or direct) would be given to the simulation model
but not to the actual process, the model results would be ck served, and
the process could be ~ ~ ~ until the cooperator is satisfied that he
knows what ccr=an~s are best to commit to the actual process. there
are possibilities for having the simulator "tract" the movement of the
actual process so that any on-lLne bests could start frog automatically
updated initial conditions. The problem of what to control manually
and what to have the computer execute by following supervisory
instruction is something that cannot he solved in general but probably
~ st be decided In each near context: the on-~ne nlanni ~ simulator
might be a way to make this happen.
On-line simulation for time-delay compensation is appropriate only
to direct control, and is not necessary for supervisory control. Here
. _
OCR for page 287
287
the ~ are set to the new arm ache actual System at the same
time. Ire It's prediction (e.g., in ~ fond of a stick figure arm
or vehicle) can be ~ on top of the actual video picture
delayed in its return from space. we rare can conserve the results
freon we ~1 i~iat~ly (before the tone delay runs its course),
thereby be much Ire confident in his move before stopping for
f~dk, and! thus save several '~v~an1 wait" Cycles. These
technics have been Astral for As of we manipu~a~r arm
(Names and Sheridan, 1984) / but not yet for the manipulator and and
controlled vehicle in c~b~naticn. ~nthe~tion of vehicles or
other objects not under we c~erator's cc~trol can be predict, e.g.,
by the curator indicatir~ an each or sever successive frames were
Sin referrers nts are, these Jews can be awed to the
predictor display. With any of these pla Hi ~ pr Diction aids, the
display can be presented four any pa It of view relative to the
manip~1ator/vehicic- a feat which ~ not possible with the actual video
Bra.
state measorement/estimation/disolav has potential where all
information about what is going an "right non' is not available in
convenient form, or where measurements are subject to bias or noise, or
tiple nets may conflict. He pi ~ to provide a best
estimate of the current situa~cion or "state" (values of key variables
which indicate where the telemanipulator end effecter is relative to
referent coordinate" or-to Vital objects of interest, what are
the joint arches are joint angle ~ocitiee;, what is the level of
energy or other critical ~es, and so oath arm display this to the
human curator ~ a way which is It and Tahoe by him for
purl of Trot. this may arson ccz~bining information from Tip e
easurm~t or ds~ca-base sautes, then Biasing this information to
the extent Cat can be done (in light of available calibration delta),
and factoring in prediction of that the state Child be based on
h~awledge of that rent iris were and that are the likely system
rinses to these items. A Crete state estimation yields a "best'
prefabs ity density distribution corer all system states. Math theory
is available on state estimation but there has been a ~ t no
application to space +~lecQeratian. Some research has shown that human
operators are unable to assimilate state information that is too
complex, and tend to simplify it for themselves by estimating averages
and throwing away the full distribution, or at least by using some
simple index of dispersicn, or in the case of joint distributions over
two or more variables; by coring only the marginal distributions,
or even s~nplifyir~ to point estimates an the ir~ent variables
(P~130~h, 1986~. Remark i~ net on how to pride the ~tor
all that can be got fmn state esthetic ark how ~ display this in a
anirqfu1 way.
Supervisory command languages must be developed especially for space
telecperators. We have a good start from industrial robot command
languages (Paul, 1981) and from the few experimental supervisory
command languages which have been develcged ~ the laboratory (Brooks,
1979; Yoerger, 1982~. We must ur~crstan] better the relative roles of
analogic instruction (positioning a control device in space, pointing,
OCR for page 288
288
demonstrating a movement) and symbolic instruction (entering strings of
alphanumeric symbols in more or less natural language to convey logic,
description, contingencies, etch. Clearly in everyday discourse we
use both analogic and symbolic coring in ccmmuni~=ting with one
another, especially ~ teaching craft ski1 Is, which seem to relate
closely to what telec~peration is. Bosch Fornication modes oust be
based In c~ni~ating with a t=ler~t. the trot usually starts
with little or no "context" about the world, which objects are which
and Here they are In space. For this reason, it is necessary to tough
objects with a designated reference port on the telec~ator, to point
faith a laser Con or otherwise to identify objects (perhaps
~ncur~ntly with giving nays or reference information sy~li~ally) ,
and to ~pecif~r~eference points on those c~jects. R ~ nt pa ~ ress in
computer linguistics can contribute much to supervisory command
language.
Voice control and f-P~h~k for all the tamps it has keen sucaesbed
as an interesting telemanipulation research topic in recent years, has
seen very little systematic research. Voice command probably has the
most promise for giving "symbolic" commands to the computer (m
contrast to the normal "analogic" or geometric isomorphic commands
which the master-slave or joystick provides). Vocal symbolic cc=mands
might be used to reset certain automatic or supervisory loops such as
grasp force, or to set control gain, master slave amplitude or force
ratio, or to guide the pan, tilt and zoos of the video cameras (Bejczy
et al., 1980~.
Aids for failure detection/identification/emergenc~, response are
particularly important since in a complex system the human Orator may
have great difficult kicking when same component has begun to fail.
This can be because He current isn't being cperated and hence there
is no abnormal variable indicated. Alternatively, if it is being
operated, the variables being presented as abnormal could have resulted
from an abnormality well upstream. Finally, the operator can simply be
averioaded. Many new failure detection/diagnosis techniques have been
developed in recent years, some of them involving Bayesian and other
statistical inference, some involving multiple cc mparisons of measure
signals to on-line models of what normal response should look like, and
so on. Failure de~ionJ~iagnos~s is a critical part of supervisory
control, where the operator depends on help from the cc mputer, but
himself plays ultimate judge. This may be a prime candidate for the
use of expert systems.
Mbta-analysis of Hhman/Computer/Teleoperator/Task interaction
Abstract theory of manipulation and mechanical "sol-using has been
surprisingly lacking. Control eng veering, as it developed through the
1940-60 period, never really coped with the complex sequential
dependencies of coordinating sensory and motor activities to perform
mechanical multi-DOF manipulation tasks. Only when industrial robot
engineers began to face up to how little they knew about how to do
assembly did the need for a theory of manipulation become evident.
OCR for page 289
289
Shadow it seems reasonable that We syntax of manipulation is
analogous to that of natural Garage (i.e., tool-action-object
corresponds to subject-verWobject, with appropriate modifiers for each
term), since both are primitive human behaviors. It then sea a smut
step to apply computational linguistics to manipulation. But little of
this has been done ~s yet.
Performance measures and assessment techniques nope to he developed
for Operation. At the moment there are essentially no accepted
standards for asserting that one telemanipulator system (of hard or
software or both) is better or worse than some other. Of course to
some extent finis is context dependent, an] the success will defend upon
8 _ e ~
spell: :1C mlSSlOn ~lltemeIltS.
~ ^
But there have got to be some generic
arm ccamr~nly accept irxlices of performance Ever which could be
used to profile the capabilities of a teleoperator vehicle/manipulator
system, Including factors of physical size, strength, speed accuracy,
repeatability, versatility, reliability, etc.
One worries beeper even
terms such as accuracy, repeatability, 1 Brevity, and so on are used in
a common way within the community. No one is asking for rigid
standardization, but some commonality across Buts and measure= appears
necessary to avoid great waste and bureaucratic chaos.
Direct experimental comparisons between astronauts performing
hands-on An End and teleoperator_` perform mg either in direct or
su~ervisorv-controlleJ fashion But be done on a much more extensive
and scientifically controlled scale, making use of both the
manipulation theory and the generic performance measures to be
develcpe5. These experiments should be performed first on the y~vand
in laboratories or neutral buoyances tanks, much as Akin (1987) Han
begun, then in space on shuttle flights (e.g., EASE experiments), and
eventually on the spare station itself.
CON~rlSIONS
A Ember of research topics have been propose, all seen as critical
for the devel~nt of needed tele~rator/~1er~botic capability for
future space station arxi relate missions. These have been presented
~ e areas of:
-
(1) Sensing (with the low ~ n goal of
t~lepresence); (2) actuation (with the long term goals or versatility
and dexterity); (3) comput=~-aiding ~ supervisory control (with the
long term Goals of providing better simulation, planning and failure
detection tools, and ted erobots which are reliable and efficient in
time and energy); (4, m eta-theory of manipulation twith the long-term
,
_ ~ ~ ~ ~ ~
goals of understanding/ evaluation/ and best relative use of both human
and machine resources)
l~lerdboticsr as much as arm other resort area for the space
,
, ~
station, has dirt rest transferability to the non~over~nt
sector for use in manufacturing, construction, mint, agriculture,
Porcine and other areas With can improve our nation's pr~ctivi~r.
OCR for page 290
290
kt;~i~
Akin, D.
1987 Ongoing ~ at Mid.
Bejczy, A. K., Brawn, J. W. and Lewis, J. L.
1980 Evaluation of "smart" sensor display:; for mllltidi~sional
precision control of Apace Thee rate manipulator.
Fires of the 16th Conference on Mortal Conch. HIT,
~ridge, M\, May 5-7.
Bejczy, A. K., Dotson, R. S. arc Me or, F. P.
1980 Mdn-mac~ine s ~ interaction In a t=1 ~ rator
environment. Proceedings of the Symposium on Voice -
Interactive Systems. DOD Human Factors Group, Dallas, TX,
Any 11-13.
Brooks, T. L.
1970 SUPERMAN:
Study of Human-Comowber Interactions.
MA: Mar.
a Svstem for supervisory ~~niPNlation and the
EM thesis, Cambridge,
Corker/ K. and Bejczy, A. K.
1985 Recent advances in Presence technology development.
Proceedings of the 22nd Space Congress.
Center, FL, April 22-25
Ferrell, W. R.
1965
1966
~dy Space
Remote manipulation with trans=Jssion delay. a:::
Transactions, Human Factors in Electronics. HFE-6, 1.
Delayed force f"C~h~ck.
October.
pp. 449-455 in Human Factors.
Ferrell, W. R. and Sheridan, T. B.
1967 Supervisory control of remote manipulation.
4(10):81-88.
Goertz, R. C. and Thompson, R. C.
1954 Electronira1 ly controlled manipulator.
Nucleom cs.
1 P:~:~:
Pp. 46-47 in
Harmon, L. D.
1982 Automated Tactile sensing. International Journal of
Robotics Research 1(2):3-32.
H"=r/ E.' ed.
1973
Remotely Manned Systems.
Institute of Technology.
Pasadena, of:
California
OCR for page 291
291
Jacobson, S.
1987 Or~goir~ research at the Ur~versi~ of Utah.
Jo, E. G. and Corliss, W. R.
1967 Teleooerators and Moan Augmentation. MESA S - 5047.
Lynch, P.
1972 A Force Reflecting Joystick. ~st~r's Thesis, Department of
Mechanical Engineering, Her.
Most, R. S.
1964 Mistrial manipulators. Scientific American 211(4):88-96.
Names, M. art Sheridan, T. B.
1984 A novel predictor for telemanipulation through a time
delay. Feedings of the Army Conference on annual
Control. Mbffett Field, CA: NOVA Is R~r=h Center.
Paul, R. P.
1981 Robot~nipulators: Pr~a~ni~and Control. Cambridge,
MA: ~ Press.
Raju, J. G.
1986 An experiments master slave manipulator system to stuffer the
fe=~ibilit~r of pperator-adjustable impedance in rate
manipulation. ~n-~6hine Systems Laboratory Memo 86-1,
Err.
P=seborough, J. B. and Sheridan, T. B.
1986 Aiding human Operators with state Estonia. ~n-~c~hme
Systems Iaboratory Report. MIT, July.
Salinity, J. K.
1986 ongoing r~r~ at the Mar AI I~boratory.
Schneider, J. and Sheridan, T. B.
1984 Art Optical Tactile Sensor for Manipulator, Robotics and
~uter-Integrated Manufacturing 1~1~:65-74.
Stric~kler, T. G.
1966 Design of an Opting Mush Sensing System for a Renote
Manipulator. EM thesis, Cambridge, Ha: MIT.
Yoerger, D.
1982 Supervisory Control of Underwater lblemanipulators: Design
and Experiment. END thesis, cartridge, MA: FIT.
Yoerg~r, D.
Personal Fornication.
Representative terms from entire chapter:
human operator
