Goal-oriented Navigation System with Artificial Intelligence based Real Time Obstacle Detection

Chang, Jew Hang (2019) Goal-oriented Navigation System with Artificial Intelligence based Real Time Obstacle Detection. Final Year Project (Bachelor), Tunku Abdul Rahman University College.

Text ChangJewHang.pdf Restricted to Registered users only Download (4MB)

Abstract

Nowadays, as the development of robotic technology getting more advance, robots are developed to implement in many fields to reduce the work load and risk of human being. The environment that require the assistance of mobile robot is usually consist of obstacles and human. Hence, the robot that is able to reach the desired destination while avoiding obstacle is very important to enable robot to handle more complex environment. The main objective of this project is to develop a mobile robot that can reach a destination within an acceptable range after avoid obstacles. The robot is supposed to be able to work in a complicated environment that consist of different type of obstacle. To achieve the objective, the researches on the localization and motion planning is carried out to find out the most suitable method to be implemented. After comparing between differences methods, odometry localization and PID controller are used in this project. The odometry is used to estimate the robot position while the PID controller is used to control the movement of the robot toward destination. These methods will be work together with Convolutional Neural Network (CNN) based obstacle detection to perform the desire application. The reason why these methods is chosen is because it is easy to implement, efficient, inexpensive and low cost. According to the result from the experiments, the odometry can function well in short distance. When the distance of the desired goal become longer, the accuracy of navigation will be affected. This is due to the systematic and non-systematic error that generated during the robot moving. The error will be cumulated over time because it uses relative positioning. However, the error can be reduced by implementing compass sensor in localization. The reading of compass sensor will be used to adjust the heading of the mobile robot instead of using the orientation calculated using odometry. From this, the odometry error is able to be reduced because the fusion of sensors will help to reduce the error from each other. After all, even though there are some error exist in the experiment, the error produced is within the acceptable range which is under 10% of total distance travelled. Hence, it can be said that the method propose is efficient and able to work in the complex environment. In conclusion, the method proposed are able to reach the desired goal within acceptable range after avoid multiple obstacles. Although there is some error produce when the robot moving, but it can be reduced effectively by implementing compass sensor as the heading sensor. As a result, it is an effective method for the navigation of robot to desired point with obstacle avoidance.