Finite-state machine; Fuzzy inference system; Machine learning; Pattern; Rule base; Supervisor

This paper deals with a design of an original approach for machine learning, which allows the rule base adaptation. This approach uses a fuzzy inference mechanism for decision making, finite-state machine for the rule base switching, and the teacher Supervisor for creating the most suitable rules for the activity (skill) which is supposed to be learned. The used fuzzy inference mechanism is the integration of LFLCore, which was developed at the Institute for Research and Applications of Fuzzy Modeling. The proposed approach of machine learning was tested in individual experiments, in which the system learns to move with its joints. How the system moves with its joints is given by patterns which are submitted before the beginning of learning. The evaluated results with possible modifications are mentioned at the end of this paper together with a formulated conclusion.