This work aims to provide some new view on the learning of the rule-based systems and explore some possibilities of the autonomous processing of knowledge. Theoretical focus is given on the reinforcement learning, learning classifier systems and fuzzy logic. The second main part of the book proposes an application of the described theory as a new form of a learning rule-based system, suitable for solving tasks in the field of robotics or other problems, where a series of actions is being searched. The proposed approach is evaluated on a set of model problems. In the conclusion, weak sides and necessary improvements are discussed. The book aims to be a small step in the whole process of building an intelligent system - it puts together some ideas in hope that it will be inspiring for others to advance state the art a little more.