An interactive model of motor learning and adapting applied to a hand perturbation task
Keywords
Loading...
Authors
Issue Date
2022-07-03
Language
en
Document type
Journal Title
Journal ISSN
Volume Title
Publisher
Title
ISSN
Volume
Issue
Startpage
Endpage
DOI
Abstract
Motor adaptation is needed to adjust to dynamic environments and requires
a robust and flexible learning mechanism. Although humans show
these abilities, this is not yet achieved in state-of-the-art robots and AI.
This is due to robot manufacturing being focused on getting to a goal using
strategies programmed by humans, while the robot should be learning these
strategies (by themselves). By learning from human biological algorithms,
we may improve robot performance. Human motor adaptation has been
shown to rely on interacting learning mechanisms, namely sensory prediction
error (SPE), reinforcement learning and strategy-based learning. This
thesis will test out whether an AI model incorporating both SPE and reinforcement
learning can successfully perform a motor adaptation task. It will
also be looked at if the weight changes indicate a different learning mechanism.
The AI model chosen was a songbird model, with the experimental
framework being a hand perturbation task. It was expected that the model
would be able to successfully learn the motor movement, adapt to the perturbations
of that movement and show a mixture of updating mechanisms.
After implementing the biological network, we first show that this network
did adapt towards the required movement, but not completely. Nevertheless,
we decided to use this partially functioning network to investigate the
adaptation. Unfortunately, the network could not adapt to the perturbations.
Due to this result, We could not infer the learning mechanism. We
cannot say if a partially working network is learning differently if the results
are not in line with what is desired. Therefore we conclude that the
learning mechanism is not different. On the other hand, interesting insights
were found in the way the weights behaved. They seemed to mimic the
trajectories found in neurons.
Description
Citation
Supervisor
Faculty
Faculteit der Sociale Wetenschappen