|M.Sc Student||Tsabari Noa|
|Subject||Autonomous Robot Navigation Using Differential GPS and|
Image Processing Techniques
|Department||Department of Civil and Environmental Engineering||Supervisor||Professor Gilad Even-Tzur|
|Full Thesis text - in Hebrew|
During the last decades robotic technologies have become a common application for household uses. One example for such application is robotic lawn mowers (RLM). Most RLM nowadays use the same mowing concept: the robot drives in straight lines and changes direction randomly either when it runs into an obstacle or when it senses a wire that is buried at the perimeter of the lawn. RLM offer very good cutting performance and save valuable time for their owners. They less noisy and emission free thus considerably more environmental friendly than man-operated lawn mowers. Despite the described advantages, the perimeter wire has few weaknesses: the wire is relatively expensive. In addition, burring the wire requires tearing the grass and digging into the ground. This study focuses on methods for improving the navigation capability of RLM by examining two methods for autonomously navigation. Both methods maintain the principle of random navigation, but suggest an alternative way of detecting the lawn boundary.
The first method uses GNSS receivers to calculate the robot position. This method requires a preliminary step of mapping the lawn's boundary. The study examined the accuracy of the location calculated from DGPS, in which the robot location is calculated relative to base receiver. The positioning accuracy required for the robotic lawn mower is a few decimeters and sometimes even better than that. The study examined the accuracy obtained from the use of code, carrier phase and smoothed-code measurements. Obtaining an accurate solution is an important feature that the solution must provide. Nevertheless, another important feature that should be taken into account is the economic aspect. The lawn mower is intended for marketing to private users, thus reducing costs plays a significant role. Therefore, the study examined the quality of the solution obtained from inexpensive measuring equipment.
The solution obtained from the combination of a low-cost receiver and antenna showed that it is not relevant for the application of robotic lawn mower due to its low accuracy. The mowing zone is characterized by concealment and multi-path, and the position accuracy is highly affected by these phenomena.
The second method that was examined in this study uses an inexpensive camera for navigating the robot. The navigation process includes lawn boundaries identification by image processing techniques. The identification was performed using a process with several steps. The lawn was separated from other areas in the image according to color characteristics. An edge detection operator separated these potential-lawn zones from other zones in the image that surely do not represent the lawn. Then, lawn edge pixels were detected with considerations of their position in the image. Finally, the robot proximity to the boundary was roughly estimated and the decision whether to change mowing direction or not was taken. The study examined the results of the method described above using images taken from real lawns layouts.
The boundary was identified correctly most of the times. However, lawns which have green boundaries (like hedgerow) were identified partly or not at all.