Optimized Fuzzy Logic Controller and Neural Network Controller-a comparative study

摘要

This work presents the use of a Mamdani Fuzzy controller and a Neural Network controller to detect and catch an object on a 2 axes (X,Y) workspace with a robot arm. The controllers use two inputs and one output for each of two controlled axes of the robot. The inputs are defined as the position errors and the derivative of the position errors regarding the setpoint position (X_(REF), Y_(REF)) of the object to catch. The outputs of the controllers drive the X and Y axes motors. The two axes motors are actuated with the objective of capturing an object on a workspace, being the object position, i.e. the setpoint position, computed by processing an image acquired with a video camera placed over the workspace. A genetic algorithm, developed in a previous work, was used to compute the characteristics of the membership functions of the Mamdani Fuzzy Controller. The robot positioning system, using the two axes transfer functions, the Fuzzy controller and the Neural Network controller were simulated in MATLAB environment being the results presented. Also, the performances of the controllers are compared regarding the setpoint tracking accuracy, the evolution of the (x,y) trajectories over time and the control effort.