You are on page 1of 20

Preprints of the Fourth IFAC

Symposium on Robot Control


19-21, 1994, Capri.
September 19-21,1994, Capri , Italy

SPACE ROBOTICS

G. HIRZINGER
German Aerospace Research Establishment (DLR)
Institute for Robotics and System Dynamics, Oberpfaffenhofen,
Germany, fax:
D-82234 Wessling, Germany, fax : +49
+49815328-1134, df28@master.df.op.dlr.de
8153 28-1134, email: dr28@master.df.op.dlr.de

Abstract. The paper tries to outline the state of the art in space robotics. It discus-
ses the technologies used in ROTEX,
ROTEX , the first remotely controlled space robot which
flew with shuttle COLUMBIA in April 93, 93 , but it also gives a review of the major
space robot projects envisioned by different nations for the next decade. The mecha-
tronics and dynamics aspects of space robots including free-flying systems are briefly
discussed . Broader attention is given to the telerobotic and teleoperational control
discussed.
loop structures, including predictive delay-compensating graphics simulation. The
paper finally tries to emphasize that by task level programming (involving" learning
(involving "learning
by showing" on a high level) future space robots will be powerful tools for scientists
and ground operators which are not robot specialists. The tele-sensor-programming
concept of ROTEX was a first step in this direction. We plead for flying a variety
of space robot systems in the near future in order to enhance experience with and
confidence in these technologies as quickly as possible.

Keywords. Space robotics, telerobotics, shared autonomy, tele-sensor-programming,


remote control

1 INTRODUCTION Real intuition and manual skill is particularly re-


quested in non-nominal situations, e.g.
e.g. when a
After four decades of manned space flight,
flight , where tape recorder has to be repaired.
repaired . Although it is
many activities have become routine, one might not clear today when a multi-fingered robot hand
forget that the space environment continues to be might be as skilled as the human hand and when
extremely hostile to human beings. They have (if ever) a robot might show up real intelligence
to be encapsulated in vehicles (for intra-vehicular and autonomy,
autonomy, it nevertheless is obvious that even
special , extremely expensive
activities IVA) or special, with today's technology and the available telero-
suits,, which protect them from the hazard of the
suits botic concepts based on close cooperation between
space environment (for extra-vehicular activities man (e.g.
(e.g. the ground operator) and machine there
EVA).
EVA) . When comparing human skills with those are many tasks in space, where robots can replace
of present-day robots of course human beings in or at least augment human activities with reduced
general are by far superior, but when comparing cost at least from a long- term perspective.
the skill of an astronaut in a clumsy space-suit
with that of the best available robot technology,
technology, Thus we are convinced that automation and ro-
then the differences are already going to disap- botics (A&R) will become one of the most at-
pear, the more if there is a remote control and mo- tractive areas in space technology, it will al-
nitoring capability on ground with arbitrarily high low for experiment-handling, inspection, mainte-
computational and human brain power. For IVA nance, assembly and servicing with a very limited
activities a robot basically would have to compare amount of highly expensive manned missions (es-
with the full human skill and mobility; however to pecially reducing dangerous extravehicular activi-
be honest, many of the manual operations to be ties).
ties) . The expectation of an extensive technology
done in a space-laboratory environment are fairly transfer from space to earth seems to be much
simple standard operations, like handling parts, more justified than in many other areas of space
opening and closing doors,
doors , pulling drawers, pus- technology.
hing buttons etc. which have to be done just by
This is one of the main
mam reasons why several ac-
stepping through extensive, written procedures.

695
tivities towards 8pace
space robotics have 8tarted
started in a big 8ucCess
success of ROTEX has given the 8pace
space robot
number of countries. A crude discrimination of community quite a stimulus. Nevertheless a few
envisaged robot systems are large arms for exter- big projects have been under preparation since a
nal servicing attached e.g.
e.g. to the space station,
station , couple of years meanwhile,
meanwhile , and they are briefly
smaller arms for internal servicing attached inside outlined in the sequel,
sequel , too.
too .
of laboratory modules and free-flying telerobot sy-
stems for assembly, inspection and repair. repair . One of
the largest projects by sure is the space station's
station 's 2.1 The space robot technology experiment
mobile servicing system (MSS) to be built by the ROTEX
Canadian space agency. NASA's own big robot
project,
project , the flight telerobotic servicer (FTS), was The experiment's main features were as follows
cut down some time ago apparently due to exces- (fig. 1):
1) :
sive development costs, but we are sure that the
use of robots will become a major issue for NASA small,, six-axis robot (working space,.."
• A small space 1
I"'oJ

again . Other remarkable mid-term


in the future again. m) flew inside a space-lab rack.rack . Its gripper
projects are the Japanese space station ""remote remote (fig. 2) was provided with a number of sen-
manipulator system" (JEM-RMS) or the Japanese sors, especially two 6-axis force-torque wrist
ETS-7 project, an experimental flight telerobotic sensors, tactile arrays,
arrays , grasping force control,
control ,
servicer for maintenance and repair of space sy- an array of 9 laser-range finders and a tiny
stems.
stems . While the shuttle manipulator arm, arm , which pair of stereo cameras to provide a stereo
has been flown with the space-shuttle a number image out of the gripper;
gripper; in addition a fixed
of times in the past (including spectacular actions pair of cameras provided a stereo image of
like the solar-max satellite rescue in 1984),
1984) , may be the robot's
robot 's working area.
area.
seen as a crane-like predecessor system for the Ca-
nadian MSS
MSS,, there had been no space robot expe- • In order to demonstrate internal and exter-
past . The European Space
rience in Europe in the past. nal servicing prototype capabilities three ba-
Agency ESA is going to prepare the use of robots sic tasks were performed (fig. 2)
2)::
in the European part COLUMBUS of the space
a) assembling a mechanical truss structure
station, replacing EVA's
EVA 's by use of the ERA large
from three identical cube-link parts
manipulator (comparable to the JEM-RMS) and
augmenting internal activities by use of the AMTS b) connecting/disconnecting an elec-
(Automatic Manipulation and Transport System). trical plug (orbit-replaceable-unit-ORU-
Apparently those nations having less background exchange using a "bayonet closure")
and history in manned spaceflight have the stron- c) grasping a floating object, representa-
gest interest in space robotics.
robotics . We are convinced tive for grasping a floating satellite
that for complex, partly autonomous robots with
extensive ground control capabilities it would be • All operational modes which we see for the
too risky to leap from zero experience to a fully coming years were verified III
in performing
operational system;
systenl; therefore we have proposed in tasks:
these tasks:
1986 the space robot technology experiment RO- teleoperation on board (astronauts
TEX,
TEX , which has meanwhile successfully flown in using stereo-TV-monitor), see fig.
fig . 20.
space (spacelab mission D2 on shuttle flight STS
teleoperation from ground (using predic-
55 from April 26 to May 6, 93). 93) . ROTEX contai-
tive computer graphics) via human ope-
ned as much sensor-based on-board autonomy as
rators and machine intelligence as well.
possible from the present state of technology, but
on the other side assumed that for many years coo- sensor-based tele-programming
tele-programming,, as well
peration between man and machine via powerful as conventional off-line-programming.
telerobotic structures will form the basis for high- The operational modes were based on a uni-
performance space robot systems operable especi- fied control approach for which we have coi-
ally from a ground station, too [1]. [1] . ned the term tele-sensor-programming.
tele-sensor-programming.
• Typical goals of the experiment were:
were:
To demonstrate the performance of
2 SPACE ROBOT PROJECTS - A BRIEF a complex, multisensory robot system
REVIEW with powerful man-machine-interfaces
(as are 3D-stereo-computergraphics, 6
Let us refer to some of the most interesting space dof control ball
ball,, stereo imaging),
imaging) , in a
robot projects mentioned in the introduction in a variety of operational modes including
little bit more detail. In general we think that the ground control.

696
To verify self-calibration and joint con- accommodation) following e.g. e.g. the resolved rate
trol under zero gravity, as well as pg-
J.lg- control structures as proposed by Whitney [18]. [18] .
motion planning based on the requi- An important feature of the arm control capabili-'
capabili- '
rement that the robot's accelerations ties is the Artificial Vision Function (AVF) which
while moving must not disturb any J.lg-
pg- processes the images from' MSS or Space Station
fromMSS
experiments nearby.
nearby. mounted cameras in real-time and outputs positi-
p08iti-
- To demonstrate and verify the use ons, orientations and velocities with respect to the
of sensorbased 6 dof-handcontrollers camera or a reference target in the field-of-view.
balls")
("control balls ") under zero gravity. The outputs may be just displayed, or they may
be used e.g. to control the SSRMS or SPDM arm
in a feedback loop for tracking a target.
target .
Local sensory feedback from the multisensory
gripper as basis for "shared autonomy control", The MSS manipulators basically will be control-
as well as predictive delay-compensating graphics led by astronauts using separate non-force reflec-
simulations were key issues in the telerobotic con- ting hand- controllers for rotations and transla-
cepts, which made the experiment a big success. tion. Ground control was not provided originally,
We will refer to this topic later on.
on . but has been included recently in at least a basic
form .
form.

2.2 The MSS for Space Station The Japanese JEM-RMS arm, supposed to per-
form external servicing at the Japanese space sta-
The Mobile Servicing System (MSS) for the Inter- tion , is comparable to the MSS large manipula-
tion,
national Space Station will probably be the first tor arm, but at its end it will optionally carry a
really operational space robotic system around the (single arm) small fine manipulator system.
system . Fi-
year 2000. It will consist of a 7 degree-of-freedom nally the European ERA arm designed for exter-
arm with a length of about 17 m (the Space Sta- nal serving at the European COLUMBUS station
tion Remote Manipulator System SSRMS) and is in a similar way self-relocatable as mentioned
a smaller dual arm robot, the Special Purpose above with the SSRMS.
SSRMS . On the other side redun-
Dexterous Manipulator (SPDM). These two ma- dancy and self-repair capabilities in the joints of
nipulator systems can work independently of each the MSS arms are unique to our knowledge but by
other, or they can work together with the SPDM sure very important for future operational space
attached to the end of the SSRMS (fig. (fig. 3).
3) . A robot systems.
mobile transporter is supposed to move the MSS
along rails on the space station truss.
truss . The cap-
ture, manipulation and berthing of large payloads
pay loads
2.3 The AMTS project of ESA
is performed by the SSRMS, while functions re-
quiring dexterous capabilities are satisfied by the
Inside the pressurized module of the COLUMBUS
COL UMBUS
dual-arm SPDM, which will play a role in the
space station ESA plans to employ an automa-
Space Station maintenance, assembly and in pay-
tic infrastructural manipulation and transport sy-
load servicing. It is capable to operate from fix-
stem AMTS supposed to release the astronauts
tures of the mobile base system, from the end of
from routine handling tasks. The system compri-
the SSRMS, or from fixtures on other structures
sing a 7 dof manipulator will be mobile through
[43].
[43] . the laboratory by use of a 3 dof translational rail
All arms are similar in kinematics having clusters system, so that in combination 10 degrees of fre-
edom are available and the robot may reach any
of 3 joints at the shoulder and near the tip, with
point in the lab. It is supposed to use an updated
a pitch joint at the elbow position near the mid-
point of each arm. The SSRMS is not only mova- version of the multisensory ROTEX gripper. An
ble by the mobile transport system along rails,
rails , but appropriate robot control system called SPAll,CO
SPARCO
by its symmetrical arrangement of joints it has a is under development presently guided by the Ita-
self-relocation capability and thus can step from lian company TECHNOSPAZIO.
one power data grapple fixture to the next as dis-
tributed around the station, thus reaching areas
which cannot be reached by the mobile transpor- 2.4 The ETS-7 project
ter. Different control modes are provided ranging
from manual single joint commands to full carte- ETS-7 is the 7th engineering test satellite of
sian control (based on selectable reference frames) NASDA (National Space Development Agency of
and to the automatic execution of definable m<r mo- Japan) to develop and perform in-orbit experi-
tion sequences. Force-torque-sensors in the wrist ment of rendezvous and docking (RVD) and space
allow to close local sensory loops (force-moment robot technologies. It presumably will be the first

697
free-flying telerobot system to be launched around exchange mechanisms are important for enhan enhan·
1997. There will be a chaser satellite with a force- ser·
cing versatility of these systems. For internal ser·
controllable 6 dof robot arm (and separate from vicing we believe that two-finger grippers are ar
it on an own base an advanced robotic hand, as adequate solution, the more as in designing spac~
developed by MITI and ETL), and a target satel- laboratories the parts to be handled (or e.g. do
lite (fig. 4). Automatic rendezvous and docking ors and closures to be opened) may be provide<
experiments between chaser and target will be with the appropriate interface that allows optiopti-
augmented by a number of robotic experiments mal force-torque transmission. Most space robo!
robol
aa are orbit replaceable unit exchange, simulated
u endeffectors envisaged will use force-torque wris1
wris!
fuel transfer, simulated battery exchange, target cameras. However sensor integration
sensors and cameras.
handling.
satellite handling. especially in free space with its temperature an<
aD<
radiation problems is fairly complicated and ex ex-
This last task would include that the robot grasps pensive.
pensive. Even inside laboratories space qualifica·
qualifica
the target (attached to the chaser), the docking tion requirements tend to hinder the use of advan·
advan
mechanism is released (the arm has to follow), and ROTE~
integration . Let us take ROTE]
ced electronics integration.
then the arm moves the target around. Capture as an example.
of target satellite is still under consideration. The
chaser is 2,2 tons and the target 0,4 tons,
tons , while the With more than 1000 tiny SMD electronic anc aD<
ORU is only 10 kg. kg. So the robot arm control sy- several hundred mechanical components the RO·
stem must be robust for payload mass variations. TEX gripper is probably the most complex mul-
Special attention is paid to the cooperative con- tisensory endeffector that has been built so far
trol of satellite attitude and robot arm (see section The gripper was not space qualifiable on compo-
Ill).
Ill) . So one of the envisioned tasks is to identify nent level especially because SMD technology ifi~
the coupled dynamics of the satellite mounted ro- not yet generally permitted in space; so it hac
bot arm and the satellite with appendages such to undergo vibration, temperature, off-gasing ancan<i
as
u the solar panel and the antenna. Other tasks so-calle~
EMC-tests as a whole; we finally got a so-callec
are to verify feedforward control techniques, where " waiver" , i.e. a special permission, from NASA
"waiver",
the estimated disturbance torques of the robot are As a positive experience during the flight all grip-
fed into the attitude control system, and to mon- per sensory and drive systems (which were per-
attitude
tion planing of the arm which minimizes .attitude manently sending their temperatures down in ad-
errors of the chaser. dition to the measuring values) remained belo"belm'!
40°C, while being operable up to 80°C. All com-
By the way in the European community compara- ponents worked absolutely reliable.
ble concepts toward free-flying robots have been
developed. The Italian SPIDER project e.g. has We think that in the future articulated multi·
a dual arm free-flying robot system in mind, while e.g. bio-
fingered hands will be useful in handling e.g.
the ESS (experimental serving satellite) studies of logical experiments or in repairing systems, but
the German space agency DARA (although in a presently all mid-term projects aim at simpler en-
very premature state) aim at a free- flying robot deffectors. Concerning space robot arms the sa-
wh·ch
which might inspect (and possibly repair) one of ving of weight is a crucial item. Even in the RO-
geoetationary satellites, which do not (or
the large geostationary TEX flight model fig. 6 (which was a comparabl~
comparably
do no longer) work; a target example is the TV- small arm with 1 m range), weight saving in the
Sat1, a solar panel of which did not open, so it
Satl, joints led to the effect, that the robot on ground
never became operable since launch in 1987. had to be Sll pended by strings for restricted test
suspended
operations. Large arms as the shuttle manipula-
tor normally are suspended by air bearings in flat
floor tests; but as an interesting fact the Cana-
3 MECHATRONICS AND DYNAMICS dian MSS arm to our knowledge will not use these
ASPECTS kind of expensive and yet restricted test facilities,
but will fully rely on computer simulatioDB
simulations stu-
Design and control of space robot mechanisms is
Deaign dies.
dies. Another interesting observation is, that even
characterized by a number of special restrictions in large arms like the shuttle manipulator, nearly
technology.
poeed by space environment and technology. all of the elasticity that calls for active damping
control (not available in the shuttle manipulator)
Endeffectors are often designed as special tools, originates from the joints, not from the arm struc-
e.g. so-called grapple fixtures (fig. 5) with own tures. Again using ROTEX experience, one has to
latching or locking mechanisms e.g. for stepping take in account that backlash in space is perma-
along space station structures, or for grasping nently active due to missing gravity preloading,
floating satellites (which of course should provide and that adaptive joint control using friction ob-
the corresponding counterpart). Automatic tool

698
servers is very helpful. Different type of friction be developed as shown in [7].
observing joint controllers were uploaded during
the ROTEX mission. • the spacecraft is not controlled (free-floating
system); of course this approach can only
We believe that space robotics may become a ma- be used in the absence of external forces
jor driver for a new,
new , extremely light-weight ro-r~ and torques. In [8] the existence of "dyna-
bot generation. For the astronaut training in RO- mic, path dependent singularities" has been
TEX we developed a kinematically identical light- shown . When they are avoided, and the cor-
shown.
weight version of the flight model, which was fully rect dynamic models are implied, then nearly
operable on ground, and where we integrated new, any terrestrial control is applicable [9], see
extremely compact joints with gear reduction 1 : [10] .
also [10].
600 into carbon fibre grid structures [2], see fig. 7. • the spacecraft's attitude and position con-
It shows up a 1:1
1: 1 ratio between own weight and
trol system is controlled in a coordinated way
load capability (10 kg both). Presently a new ver-
to reach desired locations and orientations in
sion of this arm is provided with torque-controlled space, thus realizing an unlimited workspace
joints based on inductive torque sensing; for us
with a practically redundant manipulator.
it leads into a new generation of multisensory,
torque-controlled (instead of position-controlled)
space robots. In addition to these basic control categories as
proposed in the literature there are additional ef-
In the final version all power and signal electronics fects which are very difficult to simulate, e.g. im-
will be integrated in the arm using SMD minia- pact between the satellite-robot system and a tar-
electronics. The remaining problem will
turized electronics. get to be grasped. In [11] theoretical investigati-
then be to qualify this new 7 dof arm for space, e.g. ons and airbearing flat-floor experiments are de-
be enveloping the link structures by appropriate scribed which include these complicated mecha-
foils or moving into more special hybrid electr~
electro- nIsms.
nics technology. By the way the arm is designed
nies
in a fully modular way and foldable to minimal
size for stowing it in a space-saving way.
4 TELEROBOTIC CONTROL LOOPS
A crucial issue in control of
offree-flying
free-flying robots is the 4.1
4.1 General remarks
interaction between robot arm and carrier satel-
lite, as in contrast to a big shuttle or space station teleoperati~
TOIn Sheridan, a pioneer in teleoperatio-
As Tom
the masses are in a comparable order of magni- nal systems, formulated [12], telerobots com-
tude. In such a system the satellite's attitude and bine the advantages of human remote con-
poeition in general is controlled actively by m~
position mo- trol with the autonomy of industrial robots.
mentum wheels and reaction jets (thrusters, which In the beginning of this kind of technology the
are also used to desaturate the wheels after they terms teleoperation and telemanipulation es-
have absorbed disturbance torques for a while).while) . sentially meant that an artificial antropomorphic
An excellent survey of existing control strategies arm (a teleoperator or telemanipulator) is direc-
is given in [6]. tly and remotely controlled by a human operator
just like an extension of his own arm and hand
Briefly summarizing, four main methodologies [14], [25]. Typically such a teleoperator system
have been investigated so far:
far : is supposed to work in hazardous and hostile en-
vironment [34], while the human operator in his
• the spacecraft's attitude and position remains safe and remote control station may make full use
e.g. feeding the robot's reaction for-
fixed by e.g. of his intelligence, especially when all relevant in-
ces 1 (gained by computation) into the sa- formation is sent back to him from the remote
tellite attitude position control system, i.e. worksite of the teleoperator, e.g. TV-images, for-
providing feedforward compensation. Then ces and distances as sensed by the manipulator.
terrestrial robot control techniques may be
used, but excessive fuel consumption may be More specifically if a teleoperator is prepared to
the result. repeat a task once shown by the operator and
especially if some local autonomy using sensory
• only the spacecraft's attitude is controlled by feedback is provided rendering the arm even more
feedforward compensation, not its translatio- adaptive, we prefer to use the above-mentioned
nal position; a kind of new kinematics has to term telerobot. Typically then the human ope-
lIB
tin ,hi.
thia paper the tenn forces
t'orces when referring to
rator serves as a supervisory master, not control-
wrist load of a manipulator .tands
tthee wriat forces and
atanda for force. aDd ling the telerobot's every move, but issuing gross
torques. commands refined by the robot or - in its ultimate

699
form - stating objectives which are executed vIa these joint controls would in addition compensate
the local robot sensors and computers. for the gravitational friction, inertial and coupling
forces using an inverse model (as indicated by the
In the following we will have a closer look into the compensating terms h,( q" q,) and hhm(qm,qm»
h,(q.,q.) m(qm, <tm))..
control structures of these telerobotic systems.
The operator in the ideal case would really sense
only the positional deviations caused e.g. by ex-
4.2
~. 2 Bilateral feedback concepts ternal forces and torqes (e.g. when colliding, lif-
ting loads etc.). In reality however such a system
Teleoperator systems are closely related to the provides reaction forces to the operator during any
master slave concept, i.e. more or less indepen- kind of motion due to the unavoidable positional
p08itional
dent of the special realization there exist opera- errors in such a servo system.
tional modes, where the manipulator slave arm
tries to track as precisely as possible the positional With these observations it comes clear that a sy-
(including rotational) or rate (i.e. velocity) com- stem as depicted in fig. 9 using an external wrist
mands given by a human master who uses some force-torque sensor between the slave's last joint
kind of an input device. In the early days of teleo- and its endeffector is superior and capable of over-
peration this input device (called master arm) coming these difficulties. The sensed external
was just a 1 : 1 replica of the slave arm, both f. at the slave's wrist has to be
force-torque vector f,
connected via mechanical coupling. More advan- multiplied by J~,
J;;', the transpose Jacobian of the
ced systems as presently used in nuclear power master arm to yield the joint torques T Tmm,, which
plants show up pure electrical coupling between are necessary to produce this same reaction force-
the two arms, thus allowing remote control over =
f m = f.
torque vector fm f, at the master's wrist [13].
[13] .
great distance provided that a TV-trans.mission The right hand side positional feedback of fig. 8 is
from slave worksite to the remote control station no longer necessary; systems of this kind are called
is used.
used . An example for this concept of bilateral bilateral force-reflecting master-slave systems.
position control is depicted in fig. fig . 8. By the The comments made above concerning inverse dy-
left hand side position control system the slave namic model techniques in the slave are still valid
joints (joint angle vector q,) are forced to follow here.
the master joints qm as closely as possible using
kind of a PD servo-controller with gain matrix S, Note that the scheme in fig. 9 is cartesian based
and damping matrix D. now , i.e. the kinematics of master and slave are
now,
independent, provided that the master arm shows
The gain matrix S, may be derived from a desi- up 6 degrees of freedom and has a similarly sha-
red cartesian stiffness matrix Sr"
Sz " following Salis- ped workspace (possibly down-scaled) compared
bury'8
bury 's stiffness control concept [13]: to the slave arm. The cartesian errors ~x, Lli ~x
are transformed via the inverse kinematics into
the corresponding joint (and if needed joint velo-
city) errors of the slave. 6-dof-hand controllers of
(1)
this type have been developed without and with
force-feedback (e.g. the kinesthetic handcon-
troller
troIler of Bejczy [14]). Although force-reflecting
relating slave joint errors ~q, and torques T, with hand controllers may provide a considerable per-
J, as the slave Jacobian.
J acobian. formance improvement [15], systems realized up
to now do not always show up the requested high
The right hand side control system in fig. 8 provi- fidelity; this has partly to do with high feedb-
des a safety- relevant kind of force reflection into ack sampling rates needed (1 kHz seems a rea-
the master arm. Assume that the slave (due to sonable value [16]), and with friction problems in
a master motion) moves into a wall or obstacle the hand-controller. In zero gravity (astronauts
not realized visually by the operator. The left as operators) no experience is yet available con-
hand side control system - with the positional er- cerning human's reaction to force reflection.
ror increasing - would force the slave arm to exert
increasing forces (eventually arriving at the ma-
ximum forces exertable by the slave joints) and
thus might destroy the slave arm or the environ- 4.3
{3 Local autonomy concepts
ment. The right hand side loop therefore feeds
back the positional joint errors (with a certain In all cases discussed so far if the operator can-
=
gain Sm = J~Sz,mJm
J~Sr , mJm generating a correspon- not see the slave robot directly he may make use
ding cartesian master stiffness Sr,m) into the ma- of a TV-transmission line and look at a monitor
ster arm joint motors. In an advanced system image (or better stereo TV image) instead of the

700
8ignal transmission
real scenery. But if there are signal tran8mi88ion In fact a scheme like that of fig. 11 works only if
delaY8 between master and slave
delays 8lave (e.g. when te- the slave robot has some inherent mechanical com-
leoperating a 8pace
space robot from ground) then the pliance SSRR either caused by the inevitable joint
bilateral schemes discussed so far implying the hu- compliance (leading to a position-dependent over-
man arm in the feedback graphics tend to fail. all compliance) or by a dedicated compliance in
Predictive (i.e. delay-compensating) computation the wrist (e.g. Draper Lab's well-known remote
and feedback of forces into the human arm seems RCC) . Now if we ask for the
center compliance RCC).
feasible when a perfect world model and a high- resulting compliance S, relating X Xm
m - X enll to the
X env
fidelity force reflection device is available, but is sensed force f..
f. exerted by the slave, we have to
difficult to realize in practice. fig . 12)
solve the equations (see fig.

Thus in the sequel we are addressing techniques


that do not provide any force-sensing in the hu-
man arm; these concepts are characterized by (x.. Xenll)SR =
(XI - Xenv)SR = SraXcompl
Sz.dxcompl =
local , artificially ge-
feedforward commands and local,
nerated compliance using impedance control
--------------
~~
f. ----.....-.--
f.
without a force sensor or active compliance
Srax
~
=
Sz.dx = Sz(x
Sr(xmm -- x
x.)
.. ) (3)
with local sensory feedback. Impedance control f.
means that the robot's endeffector reacts onto an
external force vector f just as a desired cartesian (4)
impedance, i.e.
i.e. [17]

After a few elementary calculations we arrive at

Mrx + Dr(xd - x) + Sr(Xd - x) = f (2)

(5)
where the mass, damping and stiffness matrices
Mzr , D rx , Sr
M Sx respectively are chosen arbitrarily and
Xd, Xd denotes a desired motion.
Xd, i.e. an extremely stiff slave robot (SR very do-
Sr, . while a
minant) would lead to the desired Sx,.
Fig. 10 shows a corresponding structure [4] and
Fig. very compliant slave would behave near to its na-
indicates that feedback to the human operator is tural stiffness. Clearly by appropriate choice of
now , the robot being locally compliant
only visual now, the stiffness matrix Sr
Sx one may generate different
with chosen impedance, so that it does not destroy compliance in different axes.
its environment when the artificial stiffness Sr
Sx is
chosen reasonably. tWb concepts presented
Let us recall that the last twb
were based on artificial slave compliance without
So far all the structures proposed are based on the and with a local force sensor feedback loop, and
advanced concept of direct torque control on the no force reflection into the human arm. Basically
joint level. This however is not state of the art even in case of active local compliance force re-
until now, so it is justified to look for other prac- flection into the human arm seems feasible, but
tical concepts, especially using the disturbance re- presumably it would be reasonable to supply force
jecting positional command interfaces which are feedback to the operator only in those directions
robots . Active com-
offered by all present day robots. which are not locally force-controlled.
force-controlled .
pliance concepts based on local feedback of wrist
force sensor data into the positional interface go Note that glove-like input devices as pure posi-
back to Whitney (e.g. [18]) , and have led to
(e.g. [18]), tional/rotational controllers and (force-reflecting)
different implementation proposals for telerobots too .
exosceletons fully fit into this framework, too.
[33]) .
(e.g. [16], [19], [20], [32], [33]).
Alternative concepts as developed at the German
The scheme proposed in [16], may be characteri- Aerospace Research Establishment (DLR) and wi-
zed by fig. 11, where the wrist forces are added dely applied in ROTEX are based on local sen-
to the positional errors between master and slave sory feedback and purely feedforward type of 6 dof
via a first order filter, again generating a cer~ain
certain handcontrollers, too, but the master input devices
cartesian stiffness Sr
Sx and damping via the time are designed in a way so that they work as rate
constant a (s here denotes the Laplace variable). and force-torque command systems simulta-
In the stationary case, e.g. when the master po- neously in contrast to the positional master arms
sition XXm
m has been moved behind a rigid sur- discussed so far.
far . Motions permitted are very small
face in the environment XXenv enll (fig. 12), we have (typically 1 - 2 mm via springs, i.e. no joints) ma-
ax = aXcompl
aXcompl so that the robot's motion stops.
stops. king the mechanical design fairly simple. DLR's

701
sensor or control ball (recently redesigned into using its local feedback exerts only the forces (may
a "SPACE MOUSE" contains an optically mea- be scaled) as given by the "teacher", thus is fully
suring 6 component force-torque sensor, the basic under his control, or may behave autonomously in
principle of which is also used in DLR's compliant predefinable sensor or position-controlled subspa-
wrist sensors [21]. ces. We are thus talking of shared control (see
[20], [32], [33]), i.e. some degrees of freedom are
The main features of the underlying more gene- directly controlled by a supervisor, while others
ral telerobotic concept are shown in fig. 13 and are controlled autonomously and locally by the
fig. 14. Rudimentary commands dx dX (the former machine.
deviations between master and slave arm) are de-
rived either from the sensor ball as forces (using The local loops in fig. 13 are (at least presently)
a sort of artificial stiffness relation dX =
dx = S; If) characterized by modest intelligence but high
or from a path generator connecting preprogram- bandwidth, while the loops involving the human
med points (dx being the difference between the band with but
operator's visual system are of lower bandwith
path generator's, i.e. "master's", position and the higher intelligence. Surely the long-term goal is
commanded "slave" robot position x eom com ).). Due to shift more and more autonomy to the robot's
to the above-mentioned artificial stiffness rela- site and to move the operator's commands to an
tion these commands are interpreted in a dual increasingly higher level. Shared local autonomy
way, i.e. in case of free robot motion they are and feedback control as explained above using the
interpreted as pure translational/rotational com- different type of gripper sensors was a key issue
mands; however if the robot senses contact with for the success of ROTEX.
ROTEX .
the environment, they are projected into the mu-
tually orthogonal "sensorcontrolled" (index f)
and" position-controlled" (index p) subspaces, fol-
lowing the constraint frame
frazne concept of Mason 4.4 Predictive control
[22]. These subspaces are generated by the robot
autonomously using a priori information about When teleoperating a robot in a spacecraft from
the relevant phase of the task and actual sensory ground or from another spacecraft so far away
information: to discern the different task phases that a relay satellite is necessary for communi-
(e.g. in a peg-in-hole or assembly task) automa- cation, the delay times are the crucial problem.
.tically.
tically. Of course the component dX dxpp projected Predictive computer graphics seems to be the only
into the position controlled subspace is used to way to overcome the main problems. Indeed pre-
feed the position controller; the component fJ f J pro- dictive 3D-stereographic simulation was another
sensor controlled subspace is eit-
jected into the sensorcontrolled key issue for the success of the ROTEX experi-
her compared with the sensed forces f,en,f .. n • to feed ment, with its typical round-trip delays of 5-7 se-
(via the robot's cartesian stiffness S R) the ortho- conds. Of course for these kind of ideas to work
gonally complementary force control law, (which the same control structures and path planners had
in fact is another position controller yielding a ve- to be realized on-board as well as in the predic-
locity xJ), or it is neglegted and replaced by some tive graphics ground station. And this in turn
nominal force vector f nom
nom to be kept constant e.g. meant that not only the robot's free motion but
in case of contour following. We prefer to talk also its sensory perception and feedback behaviour
about sensorcontrolled instead of force-controlled had to be realized in the" virtual environment" on
subspaces, as non-tactile (e.g. distance) informa- ground (fig. 15a).
tion may be interpreted as pseudo-force informa-
tion easily, the more as we are using the robot's Fig. 15 and fig. 16 are to outline that the human
positional interface anyway.
anyway. However we omit de- operator at the remote workstation handles a 6
tails as treated in [19] concerning transformations dof handcontroller
hand controller (e.g.
(e.g. control ball) by looking
between the different cartesian frames (e.g. hand at a "predicted" graphics model of the robot. The
system, inertial system etc.). control commands issued to this instantaneously
reacting robot simulator were sent to the remote
The "resulting" stiffness (e.g. when the path ge- robot as well using the time-delayed transmission
nerator serves as master) in the sense of fig. 11 links. In addition to the simulated preview the
and the corresponding derivations equs.(3 - 5) are ground-station computer and simulation system
the same as for the scheme of fig. 14 , given by may contain a model of the uplink and downlink
equ. (5). delay lines as well as a model of the actual states
of the real robot and its environment (especially
It is worth to be pointed out again that in case moving objects). The real down-link sensory and
of real human teleoperation although there is no joint data are compared with the down-link data
force feedback into the operator's arm, the robot as expected from the simulator, the errors are used
in an "observer" or "estimator" type of scheme to

702
correct the model assumptions. Fig. 17 and fig. 18 5 TOWARDS REMOTE TASK LEVEL
indicate the excellent coindidence of laser range PROGRAMMING
data in ROTEX.
From the experience we gained during ROTEX,
Overlaying graphically generated sceneries [23] the" any time" quick switching between on-line te-
and real TV
TV-images
-images is particularly useful for cali- sensor based off-line programming
leoperation and sensorbased
bration, e.g. for assuring that the computational (including later on-board execution), both based
viewpoint is identical to the real one, and that uniformly on shared control,
control , is the most efficient
the real world is close to the virtual one. Other telerobotic concept for the coming years. In RO-
than that it may be sufficient in many cases to TEX a basic version of such a unified approach,
overlay the predicted graphics with e.g. a wire called tele-sensor-programming, was demon-
frame representation of the wrist as derived from strated under the observation of a number of in-
the down-link joint data or even only an indication ternational observers.
hand-frame. This was realized in ROTEX,
of the hand-frame.
where the real i.e. delayed robot gripper's posi- There are some similarities of this concept with
tion was characterized just by a cross-bar and by other approaches, e.g. the tele-programming
two patches displaying the gripper's actual ope- concept of R.
R . Paul and his coworkers [39], [40].
[40] .
ning width.
In our approach any complex task like the dism<r
dismo-
In ROTEX there was only one exception from the unting and remounting of the bajonet cl08ure
closure is
local sensory feedback concept
concept.. It refers to image assumed to be composed of elemental moves,
processing. In the definition phase of ROTEX for which a certain constraint-frame- and een80r-
eensor-
(around 1986) no space qualifiable image proces- type-configuration holds (to be selected by the
sing hardware was available; nevertheless we took operator e.g. using a set of predefined alterna-
this as a challenge for the experiment "catching
" catching tives) . Thus automatic sensorbased path refine-
tives).
a free-floating object from ground" (fig. 15b). In ment is clearly defined during these motion primi-
contrast to contact operations as necessary in case tives, which request various definitions and prode-
of assembly we deal here with a nearly perfect cures, in particular
world model, as the dynamics of an object floa-
ting in zero g are well known. Hand-camera ·in-
- defining (or graphically demonstrating) the
formation on the free-flyer's pose (relative to the
nominal initial and goal situations (poeitioll8
(positions
gripper) was provided on ground using alterna-
or hand frames augmented by nominal een-
tive schemes; one was based on the" dynamic vi-
the "dynamic
patterns) ; of course in case of on-line te-
sory patterns);
sion approach" as given in [3], using only one of
leoperation the gross-path in between is also aleo
the two tiny hand-cameras, the other one was a
given by the operator, else it is generated la-
full stereo approach realized in a multitranspu-
ter (i.e.
(i .e. on-board) by the path planner.
ter system. In both cases the thus "measured"
object poses were compared with estimates as is- - providing the a-priori knowledge on the C-
sued by an extended Kalman filter that simula- space configuration and the type of shared
tes the up- and down-link delays as well as robot control (active compliance or using nominal
(fig. 16); this Kalman filter
and free-flyer models (fig. sensory patterns).
[5], [24], [41] predicts (and graphically displays)
the situation that will occur in the spacecraft af- - procedures for mapping sensory errors
errore into
ter the up-link delay has elapsed and thus allows positional/rotational errors (e.g. using & neu-
to close the "grasp
" grasp loop" either purely operator ral net training that allows to realize eenaor
8eD80r

controlled, or via shared control


control,, or purely auto- fusion, too).
too) .
nomously.
nomously. Fig. 19 shows photos of the TV-scene
- procedures for mapping positional/rotational
out of one of the hand cameras immediately before
errors into motion commands.
successful,, fully automatic grasping from ground
successful
despite of 6 sec round-trip delay, following the ~oal
- procedures for recognizing actual and goal
image processing approach in [3]. [3] . This automa- states, thus determining e.g. the end of an
&0
tic, ground-controlled capture of the free-flyer was move.
elemental move.
one of the many spectacular actions of ROTEX ROTEX..
evertheless we have to state that this kind of
Nevertheless
delay-compensating control structure will not be An impressive verification of these concepts was
as
typical for future space robots,
robots , as local vision feed- given during the mission when the ground opera-
back will be feasible with on-board vision systems, tor in on-line teleoperation stepped through the
too.
too . ORU-exchange task without waiting at the end
of the elemental moves until the robot in space
had confirmed reaching the goal situation of the

703
corresponding elemental move.
move. In the same way sing different levels of sensor-based robot auto-
we were able to on-line teleoperate in the virtual nomy was important for ROTEX and will be cru-
environment, but send up the gross commands at cial for future space robots. It was clearly pro-
some arbitrary time later, e.g. after finding a sa- ven that a robot system configured
con figured in this flexible
tisfactory motion. way of arbitrary and fast switching between the
most different operational modes will be a power-
But the crucial statement is, that the tele-sensor- ful tool in assisting man in future space flight pro-
programming concept with its elemental move fea- jects and it was impressively shown that even large
sensor based on-line teleoperation
tures comprises sensorbased delays can be compensated by appropriate estima-
e.g. via predictive graphics simulation (including tion and predictive pre-simulation concepts.
local active compliance) as well as a correspon-
ding off-line-programming version, which may be For ongoing telerobotics research we presently see
characterized as "sensorbased teaching by sho- the importance of two major directions for the
wing" . Hereby the robot is graphically guided next years:
through the task (off-line on ground), storing the
relevant nominal situations (graphically simula- • to bring the sensor-based learning by sho-
ted) for later (e.g. associative) recall and refe- wing programming ideas (as outlined before)
rence in the on-board execution phase, after these to such a high level that no longer a robot
data packages have been sent up to the on-board expert is needed in order to remotely execute
path planner. Indeed this mode of tele-sensor- arbitrary tasks in e.g. a robotized space- la-
programming is a form of task-oriented, impli- boratory, but that the scientific experimenter
cit, off-line-programming which tries to overcome is sufficient for working with the robot system
the well-known problems of conventional approa-
ches, especially the fact that simulated and real • if presimulated and real sensory data show
world are not identical. But instead of e.g. ca- major differences, let the system automati-
librating the robot (which is only half the story) cally find out what was wrong (the real sensor
tele-sensor-programming provides the real ro- data or the simulated ones or the real robot's
bot with simulated sensory data that refer to rela- motion or the environment model etc.).
tive positions between gripper and environment,
thus compensating for any kind of inaccuracies in Service robots (compared to industrial robots)
the absolute positions of robot and real world. presently seem to become a major driver in ro-
botics technology progress. Space service ro-
bots in particular with the requirements for
light-weight, sensor-based autonomy and powerful
6 CONCLUSION man-machine interfaces show up a lot of potenti-
potent i-
als for terrestrial spin-off (see e.g. [42]).
[42)).
We believe and we hope that the successful flight
of ROTEX, the first remotely controlled robot in
space, has encouraged the robot community to
proceed straightforward towards designing semi- REFERENCES
autonomous space robots that relieve man from
monotonous or dangerous tasks. Time and tech-
nology is mature to fly a number of different sy- [1] G. Hirzinger, B. Brunner, J. Dietrich, J. Heindl;
stems during the next decade. To proceed quickly, "ROTEX - The First Remotely Controlled
qualification standards, very conservative in gene- Robot in Space", IEEE Int. Conference on
ral, should be reviewed, so that it will possible to Robotics and Automation, May 8-13, 1994,
lower development costs, fly more systems for the San Diego, California.
same amount of money and gain broad experience [2] G. Hirzinger, A. Baader, R. Koeppe, M. Schedl,
in short time.
time . "Towards a new generation of multisensory
light-weight robots with learning capabili-
ROTEX also clearly showed that the information ties" , IFAC'93 World Congress, Sydney, Aus-
and control structures in mission control centres tralia, July 18-23, 1993.
for future space robot applications should be im- [3] D. Dickmanns, "4D-dynamic scene analy-
proved, allowing the robot operator on ground sis with integral spatio-temporal models",
permanent direct access to the different types of Fourth Int. Symposium on Roboti~
Robotics Rese-
uplinks and providing him with a continuous TV- arch, Santa Cruz, Aug.
Aug. 1987.
transmission link.
link . [4] G. Hirzinger, J. Heindl, K. Landzettel, "Con-
trol Structures in Sensor-Based Telerobotic
Close cooperation between man (astronaut as in Systems"
Systems",, ICAR 5th Int. Conf. on Advanced
fig . 20 or ground operator) and machine compri-
fig. Robotics, Pisa, June 1991.

704
[5] Christian Fagerer and Gerhard Hirzinger, "Pre- "Se08Ory feedb-
[19] G. Hirzinger, K. Landzettel, "Sensory
dictive Telerobotic Concept for Grasping a ac~ structures for robots with supervised lear-
aclt
Object" , Proc. IFAC Workshop on
Floating Object", ning" , Proceedings IEEE Conference Robo-
ning",
Spacecraft Automation and On-Board Auto- tics and Automation, S. 627-635, St. Louis,Louia,
nomous Mission Control, Darmstadt, Sept. Missiouri, 1985.
1992 "Desisn and
[20] S. Hayati, S.T. Venkataraman, "Desi«n
[6] St. Dubowsky, E. E. Papadopoul08, "The Kine- Implementation of a Robot Control Sy,ternSy.tern
matics, Dynamics, and Control of Fee-Flying with Tra.ded
Traded and Shared Control Capability" ,
and Free-Floating Space Robotic Systems", Proceedings IEEE Conference Robotic. and
IEEE Transactions on Robotics and Automa- Automation, Scottsdale, 1989.
tion (Special Issue on Space Robotics), Vol. [21] G. Hirzinger, J. Dietrich, "Multisensory r0-
9, No. 5, pp. 531-543, October 1993. sensor based path'
bots and sen80rbased path ·generation", Pro-
R . Longman, R. Lindberg, M. Zedd, "Satellite-
[7] R. ceedings IEEE Conference Robotics and Au-
mounted robot manipulators - New kine- tomation, San Francisco, 1986.
matics and reaction moment compensation" ,
[22] M.T. Mason, "Compliance and force control
Int. J. Robotics Res. vol. 6, no. 3, pp. 87-103,
for computer controlled manipulators". Pro-
Fall 1987.
ceedings IEEE Trans.
Trans. on Systems,
Systems; Man and
[8] E.
E. Papadopoulos, S. Dubowsky, "Dynamic sin- Cybernetics, Vol SMC-11, No 6, pp.418-432,
gularities in the control of free-floating space 1981.
manipulators",, ASME J .Dyn.Syst
manipulators" .Meas.,
.Dyn.Syst.Meas.,
[23] A.K. Bejczy, W.S. Kim, St.C. Venema, "The
Contr., Vo1115,
Vol115, No. 1, pp. 44-52, March 1993.
Robot: Predictive Displays for Te-
Phantom Robot:
E. Papadopoulos, S. Dubowsky, "On the nature
[9] E.
leoperation with Time Delay", Proceedinp
of control algorithms for free-floating space
IEEE Conference Robotics and Automation,
manipulators" , IEEE Trans. Robotics Auto-
Cincinnati, 1990.
mat. Vol. 7, pp. 750-758, Dec. 1991.
[10] Y. [24] G. Hirzinger, J.J . Heindl, K. Landzettel, "Pre-
Y. Umetani and K. Yoshida, "Resolved mo-
tion rate control of space manipulators with dictive and knowledge-based telerobotic con-
concepts".. Proceedings IEEE Conference
trol concepts"
J acobian matrix", IEEE Trans.
generalized Jacobian Trans.
Robotics and Automation, Scottsdale, Ari-
on Robotics and Automation,
Automation , Vol. 5, No. 3,
1989. zona, 1989.
[11] K. Yoshida, "Space Robotics Research Ac- [25] L. Conway, R. Volz, M. Walker, "Tele-
tivity with Experimental Freefloating Robot Systems: Methods and Archi-
Autonomous Systems:
Satellite (Efforts) Simulators", Third Inter- tectures for Intermingling Autonom01l8
Autonomoua and
T~lerobotic
Telerobotic Technology", Proceedings IEEE
national Symposium on Experimental Robo-
tics, Kyoto, Oct. 1993. Conference Robotics and Automation, a.. b
leigh, 1987.
[12] T.B. Sheridan, "Merging Mind and Machine",
Technology Review, 33-40, Oct. 1989. [26] P. Simkens, J.J . De Schutter, H..
H. Van Bru_el,
BrUMel,
[13] J.J. Craig, Introduction to Robotics. Addison- "Force/Torque Sensor Emulation in a Com-
Wesley Publishing Company, ISBN 0-201- pliant Motion Simulation System", Procee-
10326-5, 1986. dings of CSME Mechanical Engineerinl
Engineerin, Fo-
rum, Toronto, 1990.
[14] A.K. Bejczy, "Robots as Man-Extension Sy-
stems in Space",
Space" , Proceedings IFAC Congress. [27] G. Hirzinger, J. Dietrich, B. Brunner, "Mul-
Budapest, 1984. Proceedinp
tisensory telerobotic concepts", Proceedings
[15] B. Hannaford, "Stability and Performance IEEE Workshop on Intelligent Motion Con-
Trade-offs in Bilateral Telemanipulation",
Telemanipulation" , trol, Istanbul, 1990.
Proceedings IEEE Conference Robotics and [28] S. Hayati, T. Lee, K. T80, P. Backes , " A teet-
K. Tso, teat-
Automation,
Automation , Scottsdale, 1989. bed for a unified teleoperated- autonOlDOUB
autonomoua
[16] W.S. Kim, B. Hannaford, A.K. Bejczy, dual-arm robotic system" . Proceedings IEEE
"Shared Compliance Control for Time- Conference Robotics and Automation, Vol 2,
Delayed Telemanipulation", Proceedings IS- 1090-1095. Cincinnati, Ohio, 1990.
MCR Symposium on Measurement and Con- [29] T.B.
T .B. Sheridan, "Human supervisory controlCODtrol
trol in Robotics, Houston, Texas, 1990. of robot systems". Proceedings IEEE Confe-
[17] T. Yoshikawa, Foundations of Robotics.
Robotics. MIT rence Robotics and Automation, San Fran-
Press, ISBN 0-262-24028-9, 1990. cisco, 1986.
[18] D.E. Whitney, "Force Feedback Control of [30] A.K. Bejczy, "Smart sensors for smart hands"
Manipulator Fine Motions". Journal of Dy- presented at the AIAA-NASA Conference on
namic Systems, Masurement and Control, 91- "Smart" Remote SeD80r8,
Se08Ors, AlAA
AIAA paper No.
97, 1977. 78-1712, Hampton, Virginia, 1978.

705
[31] J.J.
J .J. Craig, M.H. Raibert, "Hybrid Posi- [43] C.P. Trudel, D.G
D.G.. Hunter, M.E. Stieber, "Con-
tion/Force Control of Manipulators"
Manipulators".. Tran- trol and Operation of Space Manipulator Sy-
AS~1E, Journal of Dynamic Sy-
saction of the ASME, stems", NATO-AGARD Lecture Series 193,
stems, Measurement and Control, Vol. 102, Advanced Guidance and Control Aspects in
(6/ 1982) 126-132.
(6/1982) Robotics, June 1994.
[32] P.G. Backes, K.S. Tso, "UMI: An Interactive
Supervisory and Shared Control System for
Telerobotics", Proceedings IEEE Conference
Robotics and Automation, Cincinnati, 1990.
[33] W.S. Kim, B. Hannaford, A.K. Bejczy, "Force telemonipulction processOt"
elemon'pulotion processor

Reflection and Shared Compliant Control in


Operating Telemanipulators with Time De-
lay" , IEEE Trans.
lay", Trans. on Robotics and Automa-
tion, Vol. 8, No.No. 2, 1992. G
R.. Lumia, "Space Robotics:
[34] R Automata '---_. . .F
.
in Unstructured Environment", Proceedings (Oom1..-'-monioulotor )

IEEE Conference Robotics and Autmoma-


tion, Scottsdale, 1989.
ground
K.. Kosuge, J. Ishikawa, K. Furuta, M.
[35] K Sakai,,
rvl. Sakai
"Control of Single-Master Multi-Slave l\la- Ma-
nipulator System Using Vim", Proceedings ROTEX
IEEE Conference Robotics and Automation,
Cinctnnati, 1990.
Cincrnnati,
[36] S. Lee, G G.. Bekey, A.K.
A.K. Bejczy, "Computer
Control of Space-Borne Teleoperators with
Sensory Feedback", Proceedings IEEE Con-
ference Robotics and Automation
Automation,, St. Louis,
Louis,
l\lissiouri,
Missiouri, 1985.
T.B.
[37] T .B. Sheridan, "Space Teleoperation through
Time Delay: Review and Prognosis", IEEE
Transactions on Robotics and Automation
Automation::
Special Issue on Space Robotics, Vol. 9, No.
pp.. 592-606, Oct. 1993.
5, pp
W.S . Kim, A.K. Bejczy, "Demonstration of
[38] W.S.
a High Fidelity Predictive/Display Techni-
que for Telerobotic Servicing in Space" , IEEE
Transactions on Robotics and Automation:
Special Issue on Space Robotics, Vol. 9, No. No.
5, pp. Oct.. 1993.
pp . 698-702 Oct
[39] J. Funda, R .P. Paul, "Efficient control of a r~
R.P. ro-
botic system for time-delayed environments" .
Proceedings of the Fifth International Con-
ference on Robotics and Automation, pages
133-137, 1989.
[40] M.R. Stein, R R.P.
.P. Paul, "Operator interac-
tion for time-delayed teleoperation with a
behaviour-based controller" , IEEE Conf. R~ Ro-
boties
botics and Automation, San Diego, 1994.
G.. Hirzinger, K.
[41] G K. Landzettel
Landzettel,, Ch
Ch.. Fagerer, "Te-
lerobotics with large time delays - the RO-
TEX experience", IROS'94IROS'94,, Int. Conf. on In-
1. Schematic representation of ROTEX
Fig. 1.
Systems, ~lunich,
telligent Robots and Systen1s, Sept..
Munich, Sept
1994.
1994 . (upper part) and integration in spacelab
(picture, courtesy of DORNIER)
[42] G. IIirzinger, Gon1bert, J. Dietrich, JJ.. Shi,
Hirzinger , B. Gombert,
"Transferring space robot technologies into
terrestrial applications" ,25th Symp~
, 25th Intern. Sympo-
si urn on Industrial Robots, Hannover, April
sium
1994..
1994

706
Fig. 2. ROTEX experiment set-up in DLR
laboratory, where the multisensory
gripper is below the three-part truss
structure just in front of the bayonet
closure (the "ORU")

Spec'-' Puf.-._
Sp.c'-ol,.."".o..
O' .... OUI
OeI1.. OU.~I.,
LIMMpuIa,..
(SPOt.4
(SPO,,!
(SSAUS)

"0..,0.,.
"o-.lO.'a
Of.,.. F.......
O,aPP'e ,........ (4)
,POOF.)
(POOFt•

...OS
Came,.

Fig. 3. Mobile Servicing Centre with SPDM


(with courtesy of NASA/CSA)

707
Orbital R lacement
Ilc:ement Unit (ORU)

Ad.load Robotic
Advanced Robotic Hand (M I1VE11..)
HInd (MrIlIE11..\

Monitor Camera
\ Monitor Camen
Camera

Dc loyab1e Truss eN AL)

Fig. 4. ETS-7 Onboard Robot Experiment Payloads


Pay loads

Fig. 5. Latching End Effector and Power Data Grapple Fixture (courtesy CSA)

708
Fig.
Fig . 6. ROTEX-flight robot built by DORNIER (courtesy of DORNIER)
6. ROTEX-ftight

Fig.
Fig . 7.
7. DLR-light-weight robot for astronaut training

709
Fig. 8. An advanced bilateral position feedback master-slave teleoperator system

\
\
\
\
\
\
\
\
" ", \
\
-------~} master
(->-----~ joints
...............
..........
~=="-,

Fig.
Fig. 9.
9. A bilateral force feedback system, when master and slave are kinematically different
1
Iteroo-TV
IItoroo-1Y .IJ monitor

d~-------~~---~
d,.------$-r------~

Fig.
Fig . 10.
10 .

Impedance control yields a compliant slave without using a force sensor, transmission delay line indicated

710
stereo-
.tereo-TV
TV I

wrist sensor

\
~ I
I
\ I
\ I
\ I
" \ I
"" \ I
----~\ I
.." I
,..."..".." ..... I
..-A==~ I
I
I
6x I
q I
I
I
I

XI

Fig. 11: Active compliance teleoperation via the positional slave interface (transmission delays indica.ted,
indicated,
master compensation not drawn)

6x
~
I
I
I
I
I
~ .... I

I
I F ig. 12. Active compliance causes the slave
Fig.
~;v: robot to be commanded stationarily
I
I somewhat "behind" a rigid surface Zen"
Zen",
in the environment, the final position
z, not beeing identical to the master
position Zm

fine path generation gross path generation


I
I pr09rommln9.
programming.
I

__+
__ --i-- __
on on
Otl menue-control
menue-control
c:ornero. board
boord I 9round
ground stereo-
s tereo- TV

~-':OPhICD
~-:o~.~

: r----
r----~
: IJ
: IJ
I
I
IJ
gross I I
commands I II
~---_..&.._ ..... -~-,
~
:

Fig. 13: Overall loop structures for the sensor-based telerobotic concept of ROTEX

711
.---_:>-----.; f + rObOt
J+ {~~~~~iC.S, 1-_______x_R___-,
dynamics,
x
R

kinematics
q
com
com I
I
I
I
I
I potentiolly
potentially
tost
I fast
force loop
I torco
I gain oprlorl
I r--- Information
I ,-and Cp
on C ,-ond tosk Info
task
I
II q"com
q"eom
I
I
I 90noroto C
generate
I project 6x f-6)($.
I into C,andC p 6x
L_ force
control x com
low
xf,com f,
position 6xp
6x p
control I-----'-----------------~
xp,c om
p,com low
low

Fig. 14: A local closed loop concept with automatic generation of force
and position controllend directions and artificial robot stiffness

spacecraft spacecraft
reol world reol world
r-----------,
r----------..,

r-~--j L+-------.
I .. sensor-based on
!i ~-----t--~r_------__,
I:_-_-_f
L_..t
L_~
r- --.

L_"""""1 path refinement board on


I
board
III: sensor
reol molion and
real motion
dolo
data
IIreol mollon and
l: sensor doto
dolo

on on
ground
ground

world model update

gross
9 fOSS
commands
commonds
~
b_----,-<~__m_comm____.Qnd'
9'oSS

slereogfophlc s imul ation of


stert09fophle simulation
stereogrophic simulation of robot. environment,
en vironment. sensory
robot, environmen t, sensory
en,":ironment. perception and polh
path refinement do' hand
6 dot hond controller
controll er
percepllon° and path refinement
percepllon 6 dot hand controller or poth
path plonner
planner
or path planner

Fig. 15: Presimulation of sensory perception and path refinement


in case of teleoperation from ground
a) local on-board sensory feedback (e.g. tactile contact)
b) sensory feedback via groundstation (grasping a free-flyer)

712
·poce
grollnd
!
----------,------
:
rr ". . I

I
""q..
O"!/o
~----------------
#/~,..-
.~
I 0#.."
I
I r---------,
r ---------l
i
• "leri IwMwltlll,.
I
I
I
I
predlcl.d m...utld oulput I
m,.,,,,,e4 o"tpt.tl
(d.loy.d)
(d .. .,.d) I
trophic.
down-link nUm.lod .' ----,
- --I
modol
oulpul II I
II I
,obol!
.con.
...... tlon
nlimolion I I
~c.:~'c:::o:.~ er
I I
rI "
.,J
II
I
II
.up.,po.iUon~
I I I
I ~
<on""
conVoI I
I1
I
••p"p ....... '-LI J...I .lIoch 1
i~ I -----r--I
.11 •• 1.
• 1 nUmot..
co,ucUon • 0'
co,rectlonl .. limot .. I
I I I I I
I I rI I I
I I L_~t~'.!'~h1~~~
II L-
: : I
IL ______________________d~'.!_d..!r!!N~~c~. _ _ _ _ _ _ _ _ JI
J ~~~X~~~~
: I
II
L ________________________________ ~~~~~
~~~~~ ___ J ~

Fig. 16. Block structure of predictive estimation


scheme

Fig.
Fig. 17.

Sensorsimulation:
Sensorsimulation: Range finder
simulation in the" virtual" workcell
the "virtual"
environment.
environment . In addition to the 5
simulated rays out of the gripper the
bars in the right lower corner indicate
the same simulated (bright) and the
corresponding real (dark) range values
as registrated by the real robot.
robot .

ORU C .. cl .o .... O.

Fig.
Fig. 18.

Correlation between presimulated (for


comparison delayed) and real sensory
data (in closed loop each) was nearly
t
11
'.07'
•. eH
l .... (·50 ... . . eJ
(·~ n ..... c)
'.07.
', 0"
(10 • • ) '.07. perfect in ROTEX.
ROTEX . These recordings of
(·50..... e J (10 • • )
ORU £ .. ehoMg_ ORU
QAU £C .. ehong.
c hono_
the four finger-range-finders pointing
"downwards" were made during
sensorbased teleoperation
teieoperation when
'I ,fi
removing from the ORU bayonet
~

. closure (fig.
(fig. 17).
17) .
I
!

!! iI
.'i

I
'.0""
1.0'" ,.ey.
• • •• •
' .e"
'.07' . . •••
'.OP.
ttlI....
... . { .·~
[ 50..... cJJ
O ... . . e (1 0 •
(10 ."• ) tt I....
l .... (·50
( . 50..... oJ
..... 0) (10 •
(la ."•)

713
Fig. 19.
Two subsequent TV-images out of one
of the hand cameras shortly before
grasping the free flyer automatically
from ground. The dark areas at the left
and right lower part are the gripper
jaws.
Jaws.

. ....

Fig. 20:
Fig. 20 : Payload specialist Hans Schlegel tele-operating the robot via 8tereO di8play and the 6 dof
stereO TV display
handcontroller (courtesy NASA)
ASA)

714

You might also like