Baneamoon, Saeed Mohammed Saeed
(2010)
Enhanced Distributed Learning Classifier System For Simulated Mobile Robot Behaviours.
PhD thesis, Universiti Sains Malaysia.
Abstract
The four basic behaviours of mobile robot are chasing, approaching, avoiding and escaping. The main problem in robotic system is in selecting the correct behaviour. The aim of this research is to overcome the behaviour selection problem. This thesis proposes methods that can overcome the problems of good behaviour selection and good behaviour deletion. It also addresses the problem of missing information, solves the problem of oscillating between correct and incorrect behaviours, and addresses the low efficiency in mapping the input to the correct behaviour. A Distributed Learning Classifier System (DLCS) consisting of five Learning Classifier Systems (LCS) with hierarchical architecture of three levels is used. An enhanced Bucket Brigade Algorithm (BBA) is developed to avoid the problem of choosing classifiers with high strength value but with incorrect behaviour. An approach that detects steady state value for calling genetic algorithm (GA) is proposed to overcome the problems of good classifiers deletion and the local minima trap. Finally, efficient solutions for covering detectors, supporting default hierarchies formation and the oscillation between correct and incorrect action are introduced to avoid performance failure, generalisation of classifiers that have the ability to cover the specific and general conditions, and loss of desirable classifiers respectively. Overall, the enhanced approaches performed well and the enhanced learning processes proposed in the current study makes robot learning more effective. The simulated robot is tested and results have shown that it performs better with the four basic behaviours. The simulated robot is also tested on many examples of a complex behaviour which is any combination of the four basic behaviours and the results have shown that it performs better with this type of behaviours as well.
Actions (login required)
 |
View Item |
![[img]](http://eprints.usm.my/style/images/fileicons/application_pdf.png)
