A multi-layered learning model

Abstract

The growth of technology is leading mankind to an increased awareness of the need for more intelligent systems. However, one of the bottlenecks in building intelligent systems is the difficulty of acquisition, testing and refinement of domain specialists' knowledge. Learning capability offers a way through this bottleneck. In this paper, we describe a general-purpose learning model for use in an unstructured environment. The proposed model exploits different learning techniques to improve the coordination, to increase task and resource allocation efficiency and to refine problem-solving skills of system elements. The utility of such system is most evident in complex domains such as 'grasping unknown objects by a dextrous hand'. An example of the proposed model is illustrated by an intelligent dextrous hand which learns to grasp unknown objects. Moreover, an expert system for grasp mode selection was implemented in a software package and an example of grasp mode generation is demonstrated. © 1993 Kluwer Academic Publishers.