Developing Concepts from Robot Behaviour by Growing Self Organizing Networks

This paper describes an approach to developing concepts from robot actions using dynamically growing self-organising networks. A robot explores the difference between objects with different properties, e.g. rollable and pushable shapes. We use an assimilation mechanism to build robot analogs of Piaget’s first sensory-motor circular reactions. Motor schemes are accumulated in a growing self organizing network according to the Interactivist grounding hypothesis. The structures which form are similar to (very primitive) Kellyan personal constructs and provide potential distinctions for the accommodation process. The purpose is to demonstrate emergence of these internal structures which indicate separable clusters of “experience”. In this way we show how grounded, pre-linguistic “concepts” might develop from a robot’s interaction with its environment.

@InProceedings{WWP07, 
 	 author =  {Willmot, Christopher and Wermter, Stefan and Panchev, Christo},  
 	 title = {Developing Concepts from Robot Behaviour by Growing Self Organizing Networks}, 
 	 booktitle = {7th International Conference on Epigenetic Robotics},
 	 number = {},
 	 volume = {},
 	 pages = {},
 	 year = {2007},
 	 month = {},
 	 publisher = {IEEE},
 	 doi = {}, 
 }