Page 377 - Sensing, Intelligence, Motion : How Robots and Humans Move in an Unstructured World

352 HUMAN PERFORMANCE IN MOTION PLANNING

Performance Criteria. Two criteria have been used to measure subjects’ per-
formance in the tasks:

1. The length of generated path (called Path)
2. The task completion time (called Time)

In the labyrinth tests the path length is the actual length of the path a subject
generates in the labyrinth. In the arm manipulator test the path length is measured
as the sum of two modulo link rotation angles in radians. Time, in seconds, is
the time it takes a subject to complete the task.

Statistical Considerations. In statistical terms, the length of path and the
completion time are dependent variables, and the test conditions, as represented
by factors and levels, are independent variables. If, for example, we want to
compare the effect of a visible scene versus invisible scene on the length of
paths produced by the subjects, then visibility is an independent variable (with
two values, visible and invisible), and the length of path is a dependent variable.
As one would expect, the dependent variables Path and Time are highly
correlated: In Experiment One the correlation coefficient between the two is
r(Path, Time) = 0.74.
A multivariate observation for a particular subject is the set of scores of
this subject in a given task; it is thus a vector. For example, for Subject 1 the
dependent variable vector (Path, Time) in Task 1 (virtual-visible-LtoR) happened
to be (59; 175).
The concept of statistical significance (see e.g., [127]) is a quantitative index
of reliability of a given result or statement, usually in terms of a variable in ques-
tion. Specifically, the significance p-level represents the probability of an error
involved in accepting an observed result (or statement) as valid, or as representa-
tive of the population. In practice, results corresponding to the significance level
p ≤ 0.05 are usually considered significant.
Put differently, p-level indicates the probability of error when rejecting some
related null hypothesis. A null hypothesis, denoted as H 0 , relates to making
a statement about the observation data—for example, when deciding whether
two sets of data came from the same population of data. If a statistical test
suggests that the null hypothesis should be rejected, say with significance level
p ≤ 0.01, we can conclude that the two samples differ significantly, or that the
variable of interest has a significant effect on the sample data. If the test results
suggest accepting the null hypothesis, we conclude that the two samples do
not differ significantly, and hence the variable of interest has no effect on the
sample data.

7.3.2 Test Protocol

The salient characteristics of the test protocol can be summarized as follows (more
details on the experiment design and test conditions can be found in Ref. 121):