Page 44 - Human Inspired Dexterity in Robotic Manipulation

40 Human Inspired Dexterity in Robotic Manipulation

of Block 4 (RI ¼ 106.86 12.96 N mm, one sample t-test, P < .001).
The retrieval of context B, however, was better than the retrieval of context
A in Block 3, as indicated by a less negative RI (significant main effect of
Context, P ¼ .003).
We hypothesized that both the negative transfer and the retrieval failure
were caused by an interference effect generated by the preceding trial blocks.
The following two conditions were designed to examine the extent to
which break duration could affect the magnitude of the interference and
retention.
Retrieval and Interference groups: Effect of time on the magnitude of interference
on retrieval trials: There are two possible explanations for the interference on
the retrieval trials: (1) learning of context B caused context A to be
unlearned, or (2) performing context B temporarily contaminated the
retrieval of A through a secondary process without erasing learned context
A. The latter scenario would indicate that subjects could be able to recall
context A if the “contamination” could be washed out by giving subjects
a break, whereas the former scenario would indicate that context A has
to be relearned regardless of the break duration before the retrieval. There-
fore, we asked subjects in the IF10, IF20, and IF60 groups to take breaks of
different durations after rotating the object at the end of Block 2. Moreover,
we asked subjects in the RT10, RT20, and RT60 groups to take breaks after
Block 1 (Fig. 3.1B). The protocols used in the IF groups were designed to
test the decay of interference from context B, whereas the protocols used in
the RT groups were designed to provide a baseline for the ability to retain
context A over time without performing the second context.
For the RT groups, we found that the duration of the break did not sig-
nificantly affect the subjects’ ability to recall the manipulation context
learned before the break (Fig. 3.3A). Subjects could recall context
A almost perfectly (RI ¼ 27.59 14.64 N mm, 12.64 20.64 N mm,
and 18.98 23.99 N mm for the RT10, RT20, and RT60 groups,
respectively). One-way ANOVA revealed no main effect of Group
(P ¼ .904). Furthermore, one-sample t-test using RI from each group con-
firmed that these RIs were not statistically different from zero (P ¼ .10, .56,
and .45 for RT10, RT20, and RT60 groups, respectively). In contrast, we
found that the retrieval of context A improved with longer break durations
for the IF groups, suggesting a weaker interference as a function of
break duration (RI ¼ 102.14 23.36 N mm, 57.76 21.82 N mm,
and 27.83 16.43 N mm for the IF10, IF20, and IF60 groups, respec-
tively; Fig. 3.3A). Together with the RI computed from the Ctrl group