Evolutionary learning of a fuzzy controller for wall-following behavior in mobile robotics

Abstract

The design of fuzzy controllers for the implementation of behaviors in mobile robotics is a complex and highly time-consuming task. The use of machine learning techniques such as evolutionary algorithms or artificial neural networks for the learning of these controllers allows to automate the design process. In this paper, the automated design of a fuzzy controller using genetic algorithms for the implementation of the wall-following behavior in a mobile robot is described. The algorithm is based on the iterative rule learning approach, and is characterized by three main points. First, learning has no restrictions neither in the number of membership functions, nor in their values. In the second place, the training set is composed of a set of examples uniformly distributed along the universe of discourse of the variables. This warrantees that the quality of the learned behavior does not depend on the environment, and also that the robot will be capable to face different situations. Finally, the trade off between the number of rules and the quality/accuracy of the controller can be adjusted selecting the value of a parameter. Once the knowledge base has been learned, a process for its reduction and tuning is applied, increasing the cooperation between rules and reducing its number.

Quick Design of Fuzzy Controllers With Good Interpretability in Mobile Robotics

“…To face this problem, a rule reduction postprocessing is usually developed. When no restriction to the interpretability is considered, the rules can be merged [6], thus generating a scatter structure where each fuzzy rule uses different fuzzy sets for each variable. On the other hand, if we want to obtain linguistic fuzzy rules with good interpretability, a selection process can be developed to obtain a subset of the original rule base [41].…”

“…Distances are measured as the minimum distance of a set of sensors (obviously, the set of sensors is different for RD and DQ). will be a weighted sum of the orientation of each sensor in the set, giving more weight to those sensors that detect closer obstacles angle (6) where is the number of sensors in the set, angle is the angle of sensor is the measured distance of this sensor, and is the maximum distance that a sensor can measure.…”

“…Due to these difficulties, the use of learning methods for the design of fuzzy controllers has been generalized [5]. There are different approaches: evolutionary algorithms [6]- [13], neural networks [14], [15], reinforcement learning [16]- [23], a combination of neural networks and evolutionary algorithms [24]- [27], etc. Evolutionary algorithms have some characteristics that make them especially useful for the design of fuzzy controllers: they are flexible to design different components of a controller, constraints can be easily included, and they let the designer decide the most adequate tradeoff between interpretability and accuracy for a specific controller.…”

“…As a result, the learned behavior is not reliable and its implementation on the real robot is not adequate. Finally, the interpretability of some of the learned controllers is poor [6], [15], [21], [25], [27]- [30] and, as a consequence, it is difficult to detect and solve errors during the operation of the controller. This paper presents a methodology to design behaviors in mobile robotics that tries to address the above mentioned drawbacks.…”

“…Some of them take a long time to learn the behaviors [6], [7], [11], [17], [21]. Others need the definition of a lot of parameters that depend on the problem and/or environment, or a partial description of a knowledge base [9], which complicates the design of new behaviors since parameters must be tuned and expert knowledge has to be acquired.…”