| Meno: | Lejla
|
|---|
| Priezvisko: | Metohajrová
|
|---|
| Názov: | Biologically inspired predictive model of proprioceptive body representations
|
|---|
| Vedúci: | prof. Ing. Igor Farkaą, Dr.
|
|---|
| Rok: | 2016
|
|---|
| Kµúčové slová: | body schema, somatosensory, iCub, predictive model, population coding
|
|---|
| Abstrakt: | The goal of this thesis is to create a simplified, biologically inspired multimodal representation
of the body (body schema). For this we use the open simulator of the
humanoid robot iCub that learns to associate two modalities of somatosensation - proprioception
and tactile sensation. The proposed solution employs a predictive model
based on a multilayer perceptron that is trained off-line in a supervised way, given
the current proprioceptive state and an action as inputs, to predict the future body
state and a potential tactile sensation. Prior to training the model, the training data
was collected from the simulated iCub that was made to perform various types of the
self-touch using its both arms. The neural network model was successfully tested for
its ability to predict touch in the corresponding proprioceptive states, when no action
was implied as input, but also to anticipate the touch during action, shortly before it
occurred.
|
|---|