Boundary Guard for a Field Robot Based on Perception and Localization : A prototype on Husqvarna robotic lawn mower

Abstract: In this thesis we have developed from scratch a perception and localization system that bounds the robot within its operation area. The module is originally designed for a robotic lawn mower, but can be adapted to other differential-drive field robots with minor changes. With upgraded processors, multiple heterogeneous sensors including a RGB-D camera, are installed and fused with adaptive Kalman filters. A state-of-art visual SLAM package is also integrated to enhance the performance. Besides, a vision module based on segmentation is developed to detect the boundary of the lawn to further prevent the robot from running out of its scope. A path-follower controller is also developed for navigation. The system is tested and evaluated in a real lawn setting. Both localization and vision modules are proved able to bound the robot with certain margins. Due to compiling issues, we cannot run the modules concurrently on the robot platform. Nevertheless, a state machine is proposed for future integration of the implemented modulesa