Development of an autonomous rover for field applications

Abstract

Agriculture is a labor-intensive industry that requires human interactions for even the most basic actions, such as spraying and weeding. Autonomous rovers can be used for many of these tasks. This research focuses on the use of depth sensors to detect the crop and navigate the field. Initial sensor testing was conducted to determine which sensor would be used on the final system. A 2D Laser Range Finder sensor (LiDAR) was chosen for its accuracy and its relatively small data sets. The LiDAR was configured to scan the crop in front of the rover. The rover corrections were determined by an on-board computer running an algorithm written in Python. The system used a PID loop to adjust motor speed and rover heading. Three different tunings were tested. The corrections were sent to a Cube Orange Auto Pilot that allowed an integration of GPS in future works. To validate the system, indoor tests were conducted using rows made from wood and windrows made of Windrow. The data from the LiDAR scans were compared to overhead images to determine the accuracy of the system. The system showed promising results as it maintained an accuracy of 15cm 95% of the time, and an accuracy of 5 cm roughly 80% of the time at 100% power (2 m/s).