|
|
1Department of Computer Science 2Department of Social and Decision Sciences
Carnegie Mellon University Carnegie Mellon University
5000 Forbes Avenue 5000 Forbes Avenue
+1 412 268 3072 +1 412 862 2799
mathan@cs.cmu.edu jeremiah.blatz@andrew.cmu.edu
Time delayed teleoperation exacts a high toll on human cognitive resources. High error rates and poor performance times are typical consequences of operating a vehicle under such conditions. This paper describes the usability effects of simple enhancements to the interface for a teleoperated lunar vehicle. Experimental results suggest that simple interface elements such as a predictive display, steering wheel, and vehicle body representation can dramatically reduce errors and task performance times during time delayed teleoperation by inexperienced lunar vehicle operators.
Predictive display, teleoperation, lunar vehicle, time delay
Carnegie Mellon University is developing a vehicle that will allow users to navigate the lunar surface from the earth. The system is intended for walk-up-and-use operation by inexperienced users. Our role has been to propose a set of usability enhancements to an existing interface design.
Teleoperation of the lunar vehicle is complicated by the five second delay resulting from round trip communications between the vehicle and the Earth. Under such conditions vehicle users must estimate the future state of the vehicle based on current display feedback and judgments about the effect of control inputs. The mental operations required to obtain such an estimate exact a high toll on the cognitive resources of users [5]. A typical result of this is a delay in performing tasks -- to cope, users adopt a move-and-wait strategy characterized by small control inputs, followed by a wait for display feedback, followed by small inputs [3]. Control errors are another consequence of time delay; studies suggest a log-linear relationship between errors and length of time delay [1].
Figure 1: Predictive display. Arrow depicts estimated position and heading in 5 seconds; lines trace predicted path
The vehicle interface design described in this report evolved as a response to complaints voiced by users of an existing vehicle interface prototype.
The existing interface displays video from a camera mounted on the lunar vehicle and uses a mouse to control speed and direction. In tests, users found this method of controlling the vehicle unnatural a problem which may become severe in the walk-up-and-use scenarios envisioned for this lunar vehicle. Users also complained of disorientation arising from a lack of representation of the vehicle body, making it difficult for users to relate the broad terrain view provided by the display to the precise vehicle dimensions, attitude and position. Vidov and Alexander [4] cite a similar effect experienced by pilots of aircraft such as the F-16 that provide a broad unobstructed view of the terrain through a bubble canopy.
On the basis of prior usability tests and user comments, our group designed a new interface combining the following features:
The predictive display has two lines tracing the vehicleís projected path and an arrow describing the vehicle's predicted position and heading in five seconds (see Figure 1). Indirect velocity information is provided by the length of the two lines. The information presented on the predictive display is based a two dimensional calculation of vehicle location and takes into account the five second time delay. Predictions are based on a combination of user control actions and vehicle velocity; a flat lunar terrain is assumed for the computation. The rendered result of the prediction is projected into three dimensions on a video display.
In order to test the efficacy of the new design, we implemented software prototypes of the original and new vehicle interfaces taking into account the five second delay. We user tested each prototype on computer simulations of two types of lunar terrain. The simulation was built using Open GL and ran on a Silicon Graphics Indigo workstation. Users consisted of five males and three females ranging in age from 18 to 38; all subjects were Carnegie Mellon University students.
Users had 5 minutes of driving practice with each interface prototype prior to performing two timed tasks. The timed tasks required subjects to navigate two courses of varying difficulty using each interface. The first course represented a relatively flat lunar terrain while the second course required navigating a winding canyon. Subjects were asked to complete the courses as rapidly as possible without colliding with terrain obstacles. Half the subjects first performed tasks with the prototype of the original interface; the other half began with the new interface.
On average, users were significantly faster completing both courses using the new interface. However, the most dramatic difference was observed in driving errors - on average, users of the new interface made eight times fewer errors than those using the original interface.Table 1: Average performance figures.
| Interface |
|
| |
| Original |
|
|
|
| New |
|
|
|
| Statistical difference |
|
|
|
We also observed differences in driving style across the two interfaces. Users tended to employ a move-and-wait strategy while navigating with the original interface. Vehicle operation under this condition was also characterized by overcompensating control actions. In contrast, the new interface allowed continuous movement of the vehicle and afforded a much finer degree of control.
Our results indicate that simple elements such as a predictive display, steering wheel, and vehicle body representation can enhance the usability of teleoperated vehicles when there is a time delay. The enhancements incorporated in the new design were tested as a package and so it is difficult to ascribe the observed results to a particular factor. However, comments by users suggest that the predictive display was responsible for the most of the effects observed. Future testing will attempt to confirm this hypothesis. We are also keen to determine time delay lengths for which the simple elements proposed in this study remain effective.
For information contact Santosh Mathan or Arn Hyndman (mathan@cs.cmu.edu and ahyndman@cs.cmu.edu).
Our thanks Sara Kiesler for guidance, inspiration, and help with editing this paper, and to the Carnegie Mellon University students who participated in our testing.