AGV with vision navigation: A study

This thesis describes the use of a vision system for navigating an AGV through a given path. Many different kinds of sensors are used nowadays for the self-navigation of mobile robots. But most sensors limit the use and flexibility of the AGV because of its dependence in outside tools that only work with their corresponding environment. It is ideal to have an AGV that can be trained and can learn to maneuver through the paths in its environment. It is this reason why the use of neural networks is gaining popularity in the use of mobile robots. A vision system or a camera is normally the sensor that is utilized. This thesis includes how the vision system may be utilized to work with neural networks in identifying and in making the decisions in navigation. This leads to an improvement in the flexibility of an AGV by enabling AGVs to function and navigate through any environment given that the AGV is made to learn the environment. This thesis will also include the results and analysis of weights obtained from training using backpropagation neural networks and also path navigating repeatability. A training data is obtained by taking pictures of possible situations by which the camera will see, and was used in the backpropagation training program to obtain weights that will be used by the AGV. The weights are used by the AGV to determine what reaction or direction it should go given what it sees from its environment. The environment the AGV navigates on is a controlled environment of similar objects and this environment is what is used in this thesis for the AGV to learn