qaesjanabi
26-02-2008, 06:38 PM
Neural Fuzzy Controller Design for Mobile Robot Navigation
by
Qais Saif Qassim
Supervised by
Dr. Subhi Aswad Mohammed
Abstract
Autonomous navigation of mobile robots is associated with the availability of external sensors to capture information from the environment to be endowed with the ability to move through corridors. This project focuses on implementing a navigation algorithm to perform a safe and collision free navigation as well as designing a control system to control the DC servo motor to implement best control performance for an autonomous vehicle by proposing a framework for designing and simulating a mobile robot control methods.
The proposed mobile robot in this project is assumed to be using two set of sensors which are a camera and a laser sensor. The camera captures images of the ceiling of the environment (corridor) and makes calibration and correction for the heading of the mobile robot, and the laser sensor detects objects in the nearby vicinity providing the orientation and the distance of the nearest obstacles. A fuzzy navigation algorithm was implemented using a combination of three different units of fuzzy logic systems; the navigation algorithm was simulated on two different corridors. The robot navigates along the environment safety and collision free.
Different controls systems were designed, PID, Fuzzy like PID, Fuzzy like PD and Neuro-Fuzzy controllers. Each was applied to the system and tested under different conditions, the Neuro-Fuzzy controller is designed by transferring the essential elements of the fuzzy logic control system into Artificial Neural Network (ANN), which are implemented by using three ANNs, which have been trained and constructed to operate exactly (functionally) as fuzzy systems. Each ANN has been trained using two different learning algorithms, Levenberg-Marquardt (LM) and backpropagation (BP). From the performed experiments, it showed that the number of epochs required to reach the error goal for LM is less than the number of epochs required for BP, even with the same number of neurons. The Neuro-Fuzzy controller provides good response to a step input controlling the DC servo motor in which is the sole core of the mobile robot.
by
Qais Saif Qassim
Supervised by
Dr. Subhi Aswad Mohammed
Abstract
Autonomous navigation of mobile robots is associated with the availability of external sensors to capture information from the environment to be endowed with the ability to move through corridors. This project focuses on implementing a navigation algorithm to perform a safe and collision free navigation as well as designing a control system to control the DC servo motor to implement best control performance for an autonomous vehicle by proposing a framework for designing and simulating a mobile robot control methods.
The proposed mobile robot in this project is assumed to be using two set of sensors which are a camera and a laser sensor. The camera captures images of the ceiling of the environment (corridor) and makes calibration and correction for the heading of the mobile robot, and the laser sensor detects objects in the nearby vicinity providing the orientation and the distance of the nearest obstacles. A fuzzy navigation algorithm was implemented using a combination of three different units of fuzzy logic systems; the navigation algorithm was simulated on two different corridors. The robot navigates along the environment safety and collision free.
Different controls systems were designed, PID, Fuzzy like PID, Fuzzy like PD and Neuro-Fuzzy controllers. Each was applied to the system and tested under different conditions, the Neuro-Fuzzy controller is designed by transferring the essential elements of the fuzzy logic control system into Artificial Neural Network (ANN), which are implemented by using three ANNs, which have been trained and constructed to operate exactly (functionally) as fuzzy systems. Each ANN has been trained using two different learning algorithms, Levenberg-Marquardt (LM) and backpropagation (BP). From the performed experiments, it showed that the number of epochs required to reach the error goal for LM is less than the number of epochs required for BP, even with the same number of neurons. The Neuro-Fuzzy controller provides good response to a step input controlling the DC servo motor in which is the sole core of the mobile robot.