Page 47 - Human Inspired Dexterity in Robotic Manipulation

Sensorimotor Learning of Dexterous Manipulation 43

Fig. 3.4 Experimental results from Rndm group. (A) T com production for the random
(Rndm). (B) Comparisons between different trials within the Rndm group. The trial
types are labeled in A. Data are averages of all subjects. *P < .05. (Modified from
Q. Fu, M. Santello, Retention and interference of learned dexterous manipulation:
interaction between multiple sensorimotor processes, J. Neurophysiol. 113 (1) (2015)
144–155.)

A; Trials 22–24 for Context B; empty, gray, and black filled circles;
Fig. 3.4A). For context A, repeated-measure ANOVA revealed a significant
main effect of Trial (P < .001; Fig. 3.4B). The posthoc t-test confirmed that
the T com exerted on preswitch trials was significantly different from the T com
exerted in context switch trials (P ¼ .001), but not from blocked trials
(P ¼ .26). Similarly, for context B, repeated-measure ANOVA revealed a
significant main effect of Trial (P ¼ .005; Fig. 3.4B). The posthoc t-test con-
firmed that the T com from preswitch trials was significantly different from the
T com exerted in context-switch trials (P ¼ .01), but not from blocked trials
(P ¼ .73). This result suggests that the interference from preceding contexts
can be established within a few trials.
Interpretation of experimental data using model simulation: Our data clearly
revealed several important features of retention and interference in the learn-
ing of dexterous manipulation tasks. Specifically, if manipulation A is
learned, learning of the second manipulation B does not erase the learned
manipulation A, but rather generates temporary interference when retrieval
of A is required. Furthermore, the first block of context A also generates a
temporary interference on the initial performance of context B. Despite
these interferences, the learned manipulation was shown to be quite robust
with respect to time and retrievable after the interference was washed out.
These findings suggest that there are at least two processes underlying
learning of dexterous manipulation. The first process updates the context-
dependent internal representation of the manipulation leading to the
trial-by-trial error reduction as well as long-term retention. The second