|
Scaling up of
lexicon and action repertoire in Cognitive Agents
Description:
Progressive
learning of basic gestures into composite actions
Previous
simulations consisted of learning actions or a combination of
actions and words. In this final simulation we are taking a step
backwards in the categorization of actions. We are breaking down
the action into basic gestures. Before learning a completed
action we are interested in the systematic breakdown of actions
into individual gestures, that is to say for example a
two-handed action would be broken down into two single
handed-actions and analyzed as individual steps in the process
of a compound action. As an extension to the previous
simulations, each feature is added dynamically. Firstly the
simulation starts with the left-handed action. At timestep 10000
(1/3rd of the simulation) we consider the right-handed action,
using the same dynamics of the fuzziness values as for
simulation 2 and finally at timestep 20000 we consider the
phonetic feature. Figure 1 shows that the model is able to
dynamically adapt to compound action associated with the word
generation.
Results:
Figure 1: Time evolution of the fields using as input the
composite action and phonetic feature: 112 different composite
actions + 112 words
Up -
Back
|
