Page 45 - Human Inspired Dexterity in Robotic Manipulation

Sensorimotor Learning of Dexterous Manipulation 41

performing context A, one-way ANOVA confirmed a significant main
effect of Group (P < .001). Moreover, one-sample t-test revealed that only
the IF10 and IF20 groups had significantly negative RI (P ¼ .003 and .033
for IF10 and IF20 groups, respectively), but not the IF60 group (P ¼ .118).
This result suggests that the learning of manipulation in the second context
B did not erase the learning of context A. Instead, the learning in context
B temporarily impaired subjects’ ability to retrieve learned context A, which
could be washed out by a 60-min break. One-way ANOVA confirmed that
the RI’s of context A from the IF60 and RT60 groups were not significantly
different from each other (P ¼ .764). Additionally, we found that the block-
ing effect on context retrieval was likely induced by the most recently
performed trials. Specifically, for the IF groups who were asked to retrieve
context B in Block 4, they all exhibited imperfect retrieval again to a similar
extent (one-way ANOVA, P ¼ .685; RI ¼ 139.42 14.71, averaged
across the three groups), despite the different break durations before Block 3.
Importantly, our data also suggest that subjects were able to well retain
the learned context regardless of the decay of the temporary interference. To
further verify the subjects’ ability to retain learned manipulation, we asked
the Ctrl group to come back after 2 weeks (IF2W) and recall the manipu-
lation tasks with the same block sequence (Fig. 3.1B). We found that sub-
jects were able to retrieve context A to a similar extent as the IF60 group
(RI ¼ 23.14 22.12 N mm; one-way ANOVA, P ¼ .874; Fig. 3.3A).
This result extends previous observations of a 24-h retention of learned
manipulation, for example, retention of anticipatory control of fingertip
forces to object mass learned through object lifting [20,24]. Additionally,
we found interference again when subjects had to recall context B in the
second session after 2 weeks (RI ¼ 181.46 21.67 N mm) due to the
most recent performance in the preceding context A.
Transfer group: Time-dependent interference on the transfer trial: We have
demonstrated that learning manipulation in context B temporarily interferes
with the retrieval of previously learned manipulation in context A. This
raises the question of whether learning context A in the first block would
exert a similar anterograde interference on the subsequent manipulation
in context B. We hypothesized that if a long enough break was given after
Block 1, subjects could have started learning context B with a smaller initial
error. In the Transfer group, subjects performed one block of eight trials and
rotated the object at the end of the block (Fig. 3.1B). After a one-hour
break, the subjects came back and started to learn the manipulation in
context B. As expected, the T com in Trial 1 of Block 2 for the TF60 group