Download as pdf or txt
Download as pdf or txt
You are on page 1of 6

Proceedings of the 2002 IEEE

InternationalConference on Robotics 8 Automation


Washington, DC May 2002

Effects of Time Delay on Telerobotic Control


of Neutral Buoyancy Vehicles

J. Corde Lane, Craig R. Carignan, Brook R. Sullivan, David L. Akin, Teresa Hunt, and Rob Cohen

Space Systems Laboratory


University of Maryland
College Park, MD 20742

Abstract
The use of telerobots in remote locations allows humans
to extend their capabilities to distant and hazardous
environments. Although safer for the operator, this
extension of capabilities comes at a price. Due to
communication across great distances, experiencing time
delays of up to several seconds is common. Any time
delay in the humadrobot interaction can significantly
degrade the effectivenessof operator control. This paper
will looks at two common instances of human-in-the-loop
command of robots: command of vehicle motion and
command of a manipulator to perform an assembly task.
Results from three studies are summarized to show the
degradation in performance caused by varying levels of Figure 1: Secondary Camera and Maneuvering Platform,
time delay. SCAMP (Foreground) monitors while Ranger
1 Introduction (Background) performs an installation

For over 25 years the Space Systems Laboratory (SSL) Research has shown how time delay in the control loop
has used neutral buoyancy to simulate the microgravity between the operator and a robot can significantly
environment of space. By modifying the buoyancy of a degrade operator performance [l] [2] [3]. Held [4] found
person or robot in water, apparent weightlessness can be that even at a time delay of 0.3 seconds, operators would
achieved. The SSL utilizes a 25-foot deep, %foot decouple their commands from the robotic system s
diameter tank to accommodate representativespace response. Ferrell[5] showed that as time delay increased,
structures and robots. Many hours of space maintenance the completion time for a two-degree of freedom (DOF)
tasks have been simulated, demonstratingassembly, manipulator-positioningtask increased proportionally.
repair, and astronaut-robotic cooperation. Black [6] showed similar results illustrating the increase
in task time with higher levels of delay for a 6DOF
The Ranger telerobot, shown in Figure 1 was designed to manipulator task. Thompson [7] shows the relationship
simulate on-orbit repair and maintenance tasks using four of time delay and difficulty of the task to degrade
manipulator arms. A grapple arm (not shown) connects performance. Task difficulty was related to the degrees of
the robot with the worksite, providing mobility for the constraint when inserting an object. For example,
vehicle to optimize its working position. The two touching a flat plate has no degrees of constraint, i.e. there
dexterous arms are then used to perform the required task. are no physical restraints that would restrict motion.
A video arm (not shown) is used to provide the remote Inserting a round peg into a long slot has two degrees of
operator with the desired view of the work area. Also constraint, preventing bolh translational and rotational
shown in Figure 1, the Secondary Camera and movement against the wall of the slot. A round peg-in-
Maneuvering Platform (SCAMP) can be flown to any hole has four constraints and a square peg has five, as the
location to provide supplemental live video. Remote only degree of freedom is parallel to the axis of the hole.
operators commanded both vehicles over Ethernet via a
fiber optic tether. Thompson found that as the degrees of constraint
increase, a linear increase in completion time occurred;
0-7803-7272-71021$17.000 2002 IEEE 2874

Authorized licensed use limited to: Oulu University. Downloaded on June 06,2022 at 13:50:28 UTC from IEEE Xplore. Restrictions apply.
this linear relationship was true at several values of time 2 Free Flight Control With Time Delay
delay. The effects of time delay were additive to the
increased constraint difficulty. Warrick [8] used Free-flight capability allows the robot to move to a
compensatory pursuit tracking task with small time delays number of locations, providing better monitoring views
for up to 0.32 seconds. Performance decreased as the and improved worksite access. This ability for a robot to
delay increased, even when the subjects were not aware of quickly and accurately move to a location is essential for
the delay. monitoring or interacting with a worksite.
Previous research has shown that at about one second of 2.1 SCAMP Tests with Simulated Delay
time delay, operators begin to switch their control strategy
from continually commanding and compensating for any The experiment conducted required two experienced
time delay to a move and wait strategy [ 5 ] . With a operators of SCAMP to fly the telerobot through a three-
several second time delay, operators will input a series of dimensional course. Figure 2 shows SCAMP and the four
commands, then wait out the time delay to observe the hula-hoops used to make the course. Under different
results before sending the next series of commands. This fixed time delays the operators were to fly SCAMP
move and wait strategy can significantly increase through the course in minimal time.
completion times of tasks.
In some cases, the time delay is not fixed, but fluctuates
with time. For example, most commands may be
received after one second. However, about every 30
seconds a single command may take 3 seconds to reach its
destination or may be dropped completely. This is
common when sending commands over the Intemet; a
variable time delay exists due to successive
communication packets being routed differently. It has
been found that a short variable time delay can be more Figure 2: The Test Course
detrimental than a longer fixed delay. One strategy held
commands for a fured time, which was longer than The course was made up of four 86 cm (34-inch) diameter
slowest time delay. A fixed, longer delay was traded for hoops suspended along the outside wall of the tank. The
the variable one. operator could take any path from one hoop to the next,
In this research, simulated time delay was modeled by however, SCAMP had to pass through the hoop in one
time stamping the commands made by the operator and direction. With SCAMP s 71 cm (28-inch) outer
saving them to a buffer. The simulation would then check diameter traversing through the hoops was not a simple
the time stamp periodically and determine if a sufficient task, Figure 3. The operators could maneuver SCAMP in
amount of delay had expired. If not, the command four degrees of freedom. SCAMP could translate in all
remained in the buffer until the time delay had passed. three directions: forwardhack, lewright, and upldown.
Then the simulation would execute the delayed command. Also the robot could turn in yaw to the lewright. The
This process continued, constantly adding and removing ability to pitch SCAMP up and down or perform a barrel
commands from the buffer. This method of generating roll was not available.
time delay was adequate for simulating a fixed time delay.
None of the quantitative studies in this work investigated
the effects of a variable time delay, which would require a
more featured algorithm.
There are numerous ways to ameliorate the effects of time
delay. Remote operators can use predictive displays that
can estimate where the robot will be when time delay has
expired [9] [ 101. Greater control can be given to the
robot, allowing it to independently perform a task with
either zero or minimal operator input [ 1 13 [3]. These
techniques have proven quite helpful, however, this paper
will focus on unmitigated time delay. The operator alone
is responsible for compensating for the time delay.

Authorized licensed use limited to: Oulu University. Downloaded on June 06,2022 at 13:50:28 UTC from IEEE Xplore. Restrictions apply.
To assist the operator, they were provided with three The analysis of variance (ANOVA) showed that time
video views, illustrated in Figure 4. Two fixed external delay had a significant effect on course completion time.
cameras provided overall situational awareness. Each No significance or any interactions were found with
camera could view three of the four hoops. The third subjects or trials. Figure 5 shows the increase in
view came from the onboard SCAMP camera. completion time due to time delay, averaged between
both operators and trials. The Duncan Range groupings,
at the 0.05 significant level are shown at the bottom of the
graph-

r
J
.g 300

0 0.5 1 1.5 2 2.5 3


GroupingAA A A A B B C
Time Delay [s]

Figure 5: Time Delay Effect on SCAMP Free Flight


No statistical significancewas found for one-second
delays and below. However at 1.5-second delay and
above a marked increase in delay occurred. This
corresponded with subjective comments, which reported
an increase in task difficulty as well as a transition to a
move, a wait strategy in controlling SCAMP. Although
completion times between 1.5 and 2 second delays were
not significantly different, 3-second delay completion
times were significantly higher.
Since the time delay increased with each successive test,
it is probable that learning may diminish the time delay
effect. This most likely occurred in the first (low time
delay) trials. The later tests had the benefit of learning
but the disadvantage of gTeater time delay. Retesting with
randomized time delay order may find a small increase in
completion times even at lower time delays.
(c)
Figure 4: Three views presented to the operator: (a) 2.2 SCAMP Operations Over the Internet
onboard camera view, (b) (c) fixed camera views In four separate tests, SCAMP, located at Marshall Space
Flight Center WSFC), was alternately controlled from a
To acquaint the operator with the task, they completed high school in Florida, the Johnson Space Center (JSC) in
two test runs at no time delay. The operators then Houston, and University of Maryland (UMD). In all cases
completed the course under each of the eight treatments SCAMP was successful performing monitoring
of time delay in the following order: 0,O.1,0.4,0.7, 1, operations,using remote control through the Internet,
1.5,2,and 3 seconds. The operators performed two trials even with the varying time delay. Figure 6 shows the
at each treatment. distribution of the remote control time delay. In all.cases,

2876

Authorized licensed use limited to: Oulu University. Downloaded on June 06,2022 at 13:50:28 UTC from IEEE Xplore. Restrictions apply.
the operators worked from satellite video that had about replacement unit (ORU) fluids box, shown in Figure 7.
250 ms delay. Two experienced operators performed multiple
changeouts under zero and three second time delay.

0 0.25 0.5 0.75 1


Time Delay [SI
Figure 7: Ranger Performing an ORU Fluids Box
Figure 6 : Time Delay Distribution for Remote Control of
Changeout
SCAMP
An ORU fluids box changeout task is composed of
When controlling from the University (UMD -> MSFC),
several steps. The trial would begin with Ranger s right
the variation of time delay was small. About 97% of the
manipulator in a starting position near the installed ORU.
delay was less than 300 ms. This would have made
The operator would then control the manipulator, moving
control easy, except the remaining 3% would range
the arm s tool tip to a H-Handle grasping fixture on the
between 1.5 - 6 seconds delay. To the operator, these
ORU. After grasping the ORU, the operator would
occasional long dropouts would appear as if the vehicle
activate the arm s tool drive which would in turn loosen
had ceased functioning. Since they occurred about once the bolt securing the ORU. Visual guides on the ORU
every two minutes, the operator could not anticipate the and its receptacle were provided to indicate when the
dropout and would get frustrated.
ORU was free. Finally the operator would move the arm
Controlling SCAMP from Houston (JSC -> MSFC) the to extract the ORU from the worksite. After completing
variation was larger. The majority of delay was below the extraction task, the operator would then reinstall the
800 ms (98%). Although the typical time delay varied ORU by performing the above process in reverse: insert
more, there were fewer long duration dropouts than when the ORU into the worksite, tum the bolt to secure the
controlled from the University. Operator comments were ORU, then release the ORU and move the arm away.
less concerned about these rare dropouts and instead
Several video cameras were available to assist the
indicated a desire for less variability even at the expense
operators to perform the task. Two fixed overall views
of average time deIay. were used to monitor gross positioning of the arm. A
3 Robotic Manipulation With Time Delay fixed close up view of the ORU receptacle allowed for
fine maneuvering of the manipulator and ORU. Finally,
The ability for a robot to fly to a particular location allows SCAMP s free flying view would typically follow the
for many monitoring and docking tasks. However, manipulator s tooltip as the task proceeded.
interacting with a worksite is essential in space
maintenance and assembly tasks. The goal of the Ranger Each of the two operators performed about four complete
project is to demonstrate the ability of a robot to perform changeouts (extractions and insertions) at zero and three
tasks similar to those of a suited astronaut. This would second time delay. Figure 8 shows the increase in
allow for a remote operator on the ground to work completion time due to time delay. Using a generalized
cooperatively with an astronaut in space to perform linear model ANOVA, it was confirmed that time delay
maintenance tasks. Clearly, communication time delay did cause a significant increase in completion time, at the
between the ground operator and the space telerobot can 0.01 level. It was found that the insertion task took
hamper this ability. significantly longer, at a 0.05 level, than the extraction
task. This was due to the increase difficulty of inserting
3.1 Manipulation Experimental Results the ORU into the worksite compared to the first step of
extraction by merely grasping the grapple fixture.
A study was performed using Ranger to perform a
changeout of a neutral buoyancy version of a space orbital

2877

Authorized licensed use limited to: Oulu University. Downloaded on June 06,2022 at 13:50:28 UTC from IEEE Xplore. Restrictions apply.
400
m
n
U

8
.3 300
, 3 sec T
w
c:
.3
z
&
200
E"
8 100 Figure 9: Overall view (Left) and right side close up view
(Right) from the Ranger Peg-In-Hole Simulation

0 The subjects tested under a range of time delays: 0, 0.5, 1,


Extraction Insertion 1.5 and 3 seconds. So that learning would not become a
Time Delay [s] factor, each subject performed about 10 hours of training
on the simulator to become experienced with the task.
Figure 8: Time Delay Effect on Telerobotic Extraction Then the subject would test each time delay treatment in
and Insertion of a Fluids Box ORU random order.
Subjective difference was also found significant, at a 0.01 The timed results are displayed in Figure 10. Each
level, as well as an interaction between subjects and task. treatment of time delay was statistically significant from
One subject had much higher completion times for the each other, at the 0.01 level. This supported the work
insertion task, while the second subject extraction and performed by Held [4] and Warrick [SI, which indicated
insertion tasks were much closer. that even small time delays could influence performance.
Also a linear trend with 0.983 correlation can be
3.2 Manipulation Simulation Results established using the averaged data points. Since this task
Several manipulator runs were also performed using a isolated the importance of time delay, the increase in
graphical simulation [12]. One study used five subjects, delay greatly influenced completion times.
each testing for several hours in total. The task was to
insert the barebolt tool from a simulated Ranger
manipulator into a hole. Using the simulation as a more
controlled environment, the effect of time delay could be
emphasized. External variables like vehicle power,
changes in lighting, and friction variability in the arm s
motor could be eliminated.
The subjects operated the same input devices and
software used to control the actual Ranger manipulators.
However, they only controlled a simulation that was
displayed on the computer monitor. When the testing
started, the arm would move to an initial position. Due to
ease the programming constraints a simpler task was used
0 Ii
0 0.5 1 1.5 2 2.5 3
for the simulation. The subject would then command the T h e Delay [SI
manipulator to drive the tip of the arm into a hole in the
box, show in Figure 9. When successfd, the simulation Figure 10: Linear Relationshipbetween Time Delay and
would reset the ann to a different initial position and Completion Time
another insertion would be attempted. A total of 32
insertions were performed on a given test. The subject The completion time increased by 213% when going from
was able to toggle between three fvred views. An overall zero to 3 second time delay. This is markedly greater
view, shown in Figure 9, and two orthogonal close up than the ORU experimental data previously presented,
views of the hole for fine positioning. Only one of the which had only a 47% increase. Part of this difference is
three views could be seen at a time. due to the elimination of environmental variables such as
lighting, views, etc. These variables cause perturbations
in the system making the task more challenging. Also
with only a few repetitions performed by each operator,
learning effects could have lowered the importance of
time delay in the ORU changeout task. Thus the task

2878

Authorized licensed use limited to: Oulu University. Downloaded on June 06,2022 at 13:50:28 UTC from IEEE Xplore. Restrictions apply.
became easier when performing the simulation, and perseverance has made the Space Systems Laboratory
emphasizing the effect of time delay. a success over the years.
Probably the most important factor was that the difference
between the two tasks made the simulation more References
susceptible to time delay. Time delay affects positioning [ 11 Conklin, J. E. Effect of Control Lags on Performance
tasks, and the simulation only used a positioning task.
of a Tracking Task. Journal of Experimental
The real ORU task not only required the arm to be Psychology, 53(4) 1957.
positioned, but also to grasp a fixture and turn a bolt.
These last two elements of the task are not affected by [2] Sheridan, Thomas B. and William R. Ferrel. Remote
time delay as much. Therefore the overall ORU Manipdative Control With Transmission Delay. IEEE
changeout task was less influenced by time delay. Transactions Human Factors in Electronics,4. 1963.
[ 3 ] Sheridan, Thomas B. Telerobotics, Automation, And
4 Conclusions and Future Work Human Supervisory Control. Cambridge: MIT Press,
1992.
Studies presented here supported previous research that
time delay causes an increase in task completion times. [4] Held R., A. Efstathiou, and M. Greene. Adaptation
However, the level of degradation was dependent on the to Displaced and DelayedVisual Feedback from the
task being performed. A three second delay caused a Hand. Journal of Experimental Psychology, 72. 1966.
132% increase in a free-flight maneuver task, only a 47% [5] Ferrell, William R. Remote Manipulation With
increase in a manipulator maintenance task, and a 213% Transmission Delay. IEEE Transactions in Human
increase using a simulated manipulator positioning task. Factors in Electronics,6( 1). 1965.
Compared to the free flight, where an increase for zero to [6] Black, J. H. Factorial Study Of Remote Manipulation
three-second time delay caused 132% increase in With Transmission Time Delay. M.S. Thesis MIT, 1971.
completion time, the manipulative completion time
increase only 47%. Therefore the free flight operations [7] Thompson, D. A. The Development Of A Six Degree
were influenced more by time delay than manipulator Of Freedom Robot Evaluation Test. Proceedings of 13th
operations. The greater effect on free flight was due to Annual Conference Manual Control. MIT, Cambridge,
the open loop control on SCAMP. With no input from the MA. 1977.
operator, SCAMP would continue to drift. This was not [8] Warrick, M. J. Effect of Transmission-Type Control
the case with Ranger s manipulator. The on-board closed Lags on Tracking Accuracy. Wright-Patterson AFB,
loop control would hold the manipulator to the last OH: Aero Medical Laboratory. 1969. NASA Tech Report
commanded position. Therefore, when the operators 5916.
transitioned to a move and wait control strategy during
the free flight task, the vehicle would move out of [9] Hashimoto, T., Thomas B. Sheridan, and M.V. Noyes.
position making time delay compensation much more Effects of Predicted Information in Teleoperation
difficult. Through Time Delay. Japenese J.Ergonomics,22(2).
1986.
Because of the influence of several factors, characterizing
the importance of time delay during realistic applications [IO] Noyes, Mark V. Superposition Of Graphics On Low
should be developed further. Although testing in Bit Rate VideoAs An Aid In Teleoperation. M.S. Thesis
controlled environments and use of simulations is MIT, 1982.
important, the effects-of delay on remote control of [ 1 11 Backes, Paul G. Supervised Autonomy for Space
realistic and complex tasks can help guide how best to Telerobotics . American Institute of Aeronautics and
ameliorate that delay. In addition, further research using Astronautics, Inc., Washington, DC, 1994
predictive displays can be used to help operators in both
manipulative [121 and free flight operations. The [ 121 Lane, J. Corde. Human Factors Optimization of
combination of predictive displays with greater autonomy VirtualEnvironment Attributesfor a Space Telerobotic
on board the robot may create an effective combination to Control Station. Dissertation University of Maryland,
alleviate the effects of time delay. College Park. 2000.

Acknowledgements
The authors would like to thank all the graduate and
undergraduate students who placed numerous hours on
these projects over the past 10 years. Their dedication

2879

Authorized licensed use limited to: Oulu University. Downloaded on June 06,2022 at 13:50:28 UTC from IEEE Xplore. Restrictions apply.

You might also like