Compare Between Flat and Hierarchical Architecture In Learning Classifier System for Robotics Controller
Lubna Zaghlul Bashir · 2014
A control system is an interconnection of components forming a system configuration that will provide a desired system response. A network of different Classifier Systems can implement the control system of an agent. The issue of architecture is therefore the problem of designing the network that best fits some various classes of robotics behaviors. The case study is system called (FTS),The system Seek For Food And Avoid Tree (FTS) built of three-classifier subsystem work together, each classifier system learns a simple behavior, the system as a whole has as its learning goal the control of robot activities. two levels hierarchical architecture was used. The hierarchical organization allows distinguish between two different learning activities: the learning of basic behavior and the learning of switch behavior. two classifier systems in hierarchical model, learn basic behavioral,(chase / avoid), they are used, to learn the simulated robot, single step movement in every direction in the environment, where the switch classifier systems learn to control the activities of basic classifier systems, they are used to learn, to choose between basic behavior using suppression as a composition mechanism to chose between two basic behaviors which represent complex behavior. The Seek for food and avoid Tree (FTS) system is repeated again using flat architecture by built of two learning classifier systems only (instead of three classifier systems) Organized in one level architecture( instead of two level architecture),In this work we compare between the flat architecture and the hierarchal architecture ,difference in performance between these two architectures is examined. Experimentally show that the use of hierarchal architecture can help to control the complexity of learning. our study shows that learning classifier system are a feasible tool to build control system .to achieve this goal we found very helpful to decompose the desired overall task into a set of simpler interacting classifiers organized in a hierarchy. these interacting behaviors were implemented as a set of classifier systems using a distributed architecture in which each classifier system runs on a different set of processors. this choice was sufficient to achieve adequate levels of performance for a variety of tasks. Results using hierarchal architecture show improvement over the flat architecture by reduce the learning complexity ,reduce time need to perform task, and signal accuracy.