Learning and Adaptation of Skills in Autonomous Physical Agents

A skills learning methodology is presented for autonomous physical agents. Adaptation of skills and learning is a fundamental part of the simple agent behaviours outlined. A general framework of skills learning is described that uses skill macros to define simple behaviours by agents that communicate, sense and act in the physical world. Programmed playfulness can be easily implemented in this framework that plays an important part in acquiring sophisticated skills. Reusability of results in learning algorithms is supported by ontology based classification of learning in skills. Ontologies provide references to object instances that enable modularization of software and easy interfacing of skills with learning algorithms.