We discuss the problem of learning fuzzy rules using Evolutionary Learning techniques, such as Genetic Algorithms and Learning Classifier Systems. We present ELF, a system able to evolve a population of fuzzy rules to obtain a sub-optimal Fuzzy Logic Controller. ELF tackles some of the problems typical of the Evolutionary Learning approach: competition and cooperation between fuzzy rules, evolution of general fuzzy rules, imperfect reinforcement programs, fast evolution for real-time applications, dynamic evolution of the focus of the search. We also present some of the results obtained from the application of ELF to the development of Fuzzy Logic Controllers for autonomous agents and for the classical cart-pole problem. INTRODUCTION Genetic Algorithms (GAs)[13] and Learning Classifier Systems (LCS)[7][8] emerged in the last years as powerful Evolutionary Learning (EL) techniques to identify systems that optimize some cost function. The cost function provides a reinforcement that gui...

We developed a system learning behaviors represented as sets of fuzzy rules for autonomous agents. In the past, we adopted our approach to learn successfully simple reactive behaviors, also in those cases when the evaluation function used in our reinforcement learning schema judges unevenly the different situations the autonomous agents operate on. In this paper we present a new version of our approach that can learn to coordinate many different behaviors organized in classes of mutually exclusive behaviors. The present version of our algorithm gives satisfactory results also on this new task. 1. Introduction In (Bonarini, 1993), (Bonarini, 1994a), (Bonarini, 1994b) we presented ELF (Evolutionary Learning for Fuzzy rules) a Reinforcement Learning approach we used to generate behaviors for autonomous agents. Our behaviors are sets of fuzzy rules. Therefore they have the desirable features Fuzzy Logic Controllers have, such that smoothness in the output, robustness, and so on. Up to no...

Fuzzy rules cooperate in a Fuzzy Logic Controller (FLC) to produce the best action for a given situation. If we have a population of fuzzy rules controlling a device, and we would like to evolve the population to obtain optimal performance by Reinforcement Learning, rules should compete each other, since we would like to judge their proposals. Therefore, in this approach, cooperation and competition are two opposite, desired activities done by the population members. This may be a problem, if we consider that the evaluation function may be biased, as it may happen, for instance, when we are designing a controlled device such as an Autonomous Agent. The problem becomes even harder if we would like to learn general rules, i.e., rules containing don't care symbols in their antecedents, thus competing with many groups of other rules, in many different situations. In the paper we discuss these issues, and we present our solution, implemented in ELF (Evolutionary Learning of Fuzzy rules). We...

. In this paper, we discuss situations arising with reinforcement learning algorithms, when the reinforcement is delayed. The decision to consider delayed reinforcement is typical in many applications, and we discuss some motivations for it. Then, we summarize Q-Learning, a popular algorithm to deal with delayed reinforcement, and its recent extensions to use it to learn fuzzy logic structures (Fuzzy Q-Learning). Moreover, we present how a reinforcement learning algorithm we have developed in the past (ELF - Evolutionary Learning of Fuzzy rules) implements an extension of the popular Q-Learning algorithm for the distribution of delayed reinforcement when the controller to be learnt is a Fuzzy Logic Controller (FLC). Finally, we present some examples of the application of ELF to learning FLCs that implement behaviors for an autonomous agent. Keywords. Fuzzy Logic Controllers, Q-Learning, Reinforcement Learning. 1. Introduction The main goal of the research presented in this paper i...

this paper, we present Behavioral Engineering (BE) issues, focusing on the role of learning as a support to this new branch of engineering. We discuss issues related to learn behaviors as FLC, and we propose our approach implemented in ELF (Evolutionary Learning for Fuzzy rules). We are using ELF to support the development of different types of agents. We also discuss architectural issues to combine behaviors. Finally, we present the results we obtained both in simulated and real environments.

this paper, we present issues related to learn behaviors implemented as FLCs, and we propose our approach implemented in ELF (Evolutionary Learning for Fuzzy rules). We are using ELF to support the development of different types of agents. Finally, we present the results that we have obtained both in simulated and real environments.