Comparing How Differently Embodied Agents Develop Through Perception-Action Learning
Keywords
Loading...
Authors
Issue Date
2023-07-01
Language
en
Document type
Journal Title
Journal ISSN
Volume Title
Publisher
Title
ISSN
Volume
Issue
Startpage
Endpage
DOI
Abstract
Embodied cognition is a theory that shines a spotlight on the importance of an agent
existing in an environment and possessing a body. There is significant debate over the
roles of these elements in different areas of research and in different aspects of cognition.
In this research, the focus is on exploring the role of the physical body in the
cognitive development of an agent. This is done by simulating different agents in a 2D
environment and allowing them to learn through perception-action learning. This is
a relatively new learning mechanism that ties the body’s abilities and what the agent
can learn closely together. As this mechanism is fairly new, this research also aims
to contribute to the existing methods of how to implement this mechanism. Using
elements from both reinforcement learning and curiosity, a simple and generalizable implementation
of perception-action learning is created and used to train multiple agents.
The simple implementation results in the expected behaviours of an agent learning via
perception-action learning: learning meaningful behaviours without any oversight and
having those behaviours increase in complexity over time. The resulting behaviours are
also impacted by the body types of the agents. Depending on the type of visual perception
the agent has, the agent either does not learn to manipulate the objects in its
environment, learns to manipulate one object at a time, or learns to manipulate multiple
at a time. Predictions of the variance of behaviour are non-obvious when looking at
the body type alone, demonstrating how even very simple bodies and environments can
result in interesting interactions with an agent’s cognitive development.
Keywords Embodied Cognition, Perception-Action Learning
Description
Citation
Supervisor
Faculty
Faculteit der Sociale Wetenschappen