Page 410 - Sensing, Intelligence, Motion : How Robots and Humans Move in an Unstructured World

DISCUSSION 385

one moves the arm or how continuous its motion is. If the subject stops to think
how to proceed, this in itself will not increase the path length, but it will increase
the time to task completion. The harder the task, the more thinking the operator
needs, and the more time he or she takes to think. The relation of task hardness to
the length of path is hence more subtle than its relation to the time to completion.
This may be a useful consideration for balancing advantages and disadvantages
of virtual versus physical control means in real-world teleoperation systems.

Effects of the Visibility Factor. This factor refers to the obstacles in the environ-
ment being visible or invisible to the subjects during the test. The test subjects
themselves, researchers, and practicing operators usually think that seeing the
robot surroundings would significantly improve their performance.
Interestingly, our study suggests that while this common sense judgment
applies to very easy tasks, it does not apply to relatively complex tasks. For
the easier task in this study (which is moving the arm left-to-right), there is only
a slight difference in the resulting path length and completion time. That is, see-
ing the environment helped a little in path length and in completion time. On the
other hand, for the more difficult task (moving the arm right-to-left,) there was
almost no difference in the path length and completion time. This looks puzzling,
but becomes less so if one considers that many studies have demonstrated that
humans are, in general, not very good in spatial reasoning based on visual data.
This fact questions the large resources that are often allocated in telerobotics to
help the operator see the scene. It also implies that the operator performance is
affected less by the visibility factor than by the human spatial reasoning abilities.

Effects of the Training Factor. This factor has two components that refer to the
day of the task execution: day 1, before training, and day 2, after training. When
comparing human performance on those two days, with the other conditions
fixed, any statistically significant difference should be attributed to the effect
of training. Namely, a significant difference would support a common wisdom
hypothesis that one’s performance should improve significantly after learning
from repeated exercise.
This study shows that in arm manipulator motion planning tasks, training has
no significant effect on human performance, neither in terms of path length nor
in the task completion time. In our tasks the subjects were unable to seriously
improve their motion planning skills via training. This is no doubt very surpris-
ing. One would expect the opposite conclusion: We all know examples of tasks
involving motion where, given enough training, humans become extremely adept;
an acrobat on the trapeze is but one example.
There is a big difference, however: The acrobat does a once-and-for-all learned
motion, whereas our tasks require constant spatial reasoning. Our test protocols
do not allow a subject to simply memorize a task. We want our subjects to learn
how to do a class of tasks; we want them to improve their spatial reasoning skills,
rather than memorize a specific motion. Examples of positive effect of training in
tasks that involve spatial reasoning are harder to come up with. Note that since