The University of Edinburgh -
Division of Informatics
Forrest Hill & 80 South Bridge

MSc Thesis #94133

Title:Instrumental Conditioning of a Mobile Robot: the Use of the Halperin Neuro-Connector Model for Robot Control
Date: 1994
Abstract:Halperin (1990) has described a connectionist model of learning, originally designed to elucidate social learning mechanisms in Siamese fighting fish. Halperin incorporates two basic principles which distinguish it from other connectionist models. First, she uses a variation of the usual Hebbian synaptic strengthening rule which says that it is the termination of firing in the pre- and post-synaptic neurons which is important to weight change. Second, she proposes a period of consolidation of learning: The consolidation of a strengthening event can be disrupted by events at the same synapse during a time window after the original event, thus only the last in a series of closely-spaced events will be consolidated. According to Halperin, this strengthening rule, when combined with a simple network architecture, can produce both instrumental and classical conditioning as well as various other learning phenomena such as backward (postponed) conditioning and priming. A model which produces learning has potential use as a robot controller, especially for robots which are not designed for a specific task and thus must be flexible and adaptable. A preliminary version of this model was implemented on the mobile robot Ben Hope (Hallam, Halperin and Hallam, 1993) and produced postponed conditioning. The goal of this project was to provide a robust version of the model for use with Ben Hope as well as to incorporate the new blackboard system for communication with the robot's sensors and effectors. Once the model was implemented, ten experiments testing various forms of learning were performed. The results of these experiments indicate that behaviours resembling basic phenomena involved with instrumental conditioning can be demonstrated by implementing this model on a mobile robot. Implications of this finding for both robot control and psychology are discussed.

[Search These Pages] [DAI Home Page] [Comment]