Page 49 - Human Inspired Dexterity in Robotic Manipulation

Sensorimotor Learning of Dexterous Manipulation 45

Fig. 3.5 Simulation results with the dual-processes nonlinear interaction (DPNI) model.
(A). Model output (circles) for the Ctrl group with block ABAB paradigm. (B) Model output
for the Rndm group with pseudorandom context sequence. (C) output of model
simulation for temporal characteristics of retention and interference (with 95%
confidence intervals). (D) Output of model simulation for transfer after a 1-h break
(TF60 group). (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.)

model predicts the 32-trial data averaged across subjects from the Ctrl and
Rndm group well (for the Ctrl group, compare Fig. 3.5A with Fig. 3.2,
r ¼ .95; for the Rndm group, compare Fig. 3.5B with Fig. 3.4A,
r ¼ .93). Most importantly, the DPNI model predicted the differential
effects of break duration on the phenomena of interference and retention of
learned manipulation reported here (Fig. 3.5C vs. Fig. 3.3A), as well as the
time-dependent interference on the transfer trial (Fig. 3.5D vs. Fig. 3.3B).
We found that subjects could update the internal representation of the
manipulation rapidly as indicated by the relatively large value of the learning
rate B compared to those found in reaching studies. This is consistent with
previous findings that demonstrated fast adaptation rates for learning object
manipulation when contextual cues are available [13]. The parameter C sug-
gests that the use-dependent memory u is heavily dependent on manipula-
tions performed in the most recent trials (Fig. 3.5A and B). This result is
consistent with the fast establishment of use-dependent memory observed
in different manipulation tasks [38–40], but differs from the finding that
use-dependent bias was built through repeated reaching tasks with a much