Intelligent motion planning of a mobile robot by using convolutional neural network / Siti Asmah Abdullah

• Main Author: Siti Asmah, Abdullah
• Format: Thesis
• Published: 2019
• Subjects: TK Electrical engineering. Electronics Nuclear engineering
• Online Access: http://studentsrepo.um.edu.my/11902/1/Siti_Asmah_Abdullah.jpg; http://studentsrepo.um.edu.my/11902/8/asmah.pdf; http://studentsrepo.um.edu.my/11902/

Optimizing the cost in every aspect is important in the implementation of artificial intelligence (A.I.) without affecting the accuracy of the output. Especially when it related to wastage of energy towards environment treats and resources cost. Planning the path of the mobile robot by using Convolutional Neural Network (CNN) is important in optimizing the cost in terms of energy, time and human intervention force. The robot should be able to decide the correct and safe movement depending on its current position, path reference and obstacles’ location. The mobile robot should be continuously predicting the correct movement until it reaches the targeted location by avoiding all the obstacles along its way. The intelligence of the mobile robot is imitating CNN where the network consists of Convolutional Layer, Normalization Layer, ReLU layer, Fullyconnected Layer, Softmax Layer and the last layer is Classification Layer. The network needs to undergo training session before handling the mobile robot movement. It trains by using labelled data which comes by eight different folders represent the eight different movements; up, up-right, right, right-down, down-left, left, and left-up. By the path reference created by A* algorithm the robot is capable in optimizing its path to reach the designated destination. The mobile robot is being tested in three different environment maps which come with unique and different levels of difficulty. Every map contains of static obstacle arranged in the horizontal, vertical and diagonal manners. The robot should be able to avoid the obstacle during in its path, but if the collision happens, the robot should start from its initial position, and the CNN requires to re-train to avoid the same looping decision is made. Capturing the current position of the mobile robot is very important in determining the next best move to be taken. Windowing Grid is proposed to iv solve this matter. Several algorithms are used to create the Windowing Grid in capturing the informative and relevant current position of the mobile robot. This information will be used by CNN to predict the best move as the process is repeated until the robot reaches the target position.