Hugo

Our earlier robots have not solved the Field Robot Event’s row-following problem with a sufficient degree of robustness. The objective of the work presented here was to build a robot that can detect rows consisting of small or large maize plants by using a camera system; and to provide this robot with robust localization and navigation by using probabil-istic methods to process the data from the vision system in conjunction with data from other sensors. We employed a particle filter approach where information from the robot’s wheel encoders and a gyroscope is used in the control step and where the filter is updated using information from a downward-looking camera and a laser scanner. For the weed detection and control tasks, the robot is equipped with a self-contained spray unit consist-ing of two CMUCAM3 camera’s and four narrow-cone nozzles (two on each side of the robot) which allow for precision-treatment of small areas. At the Field Robot Event, the robot was able to follow rows and turn into the correct new row in all tasks. No manual intervention was necessary; the first objective was met. In the days and weeks leading up to the event, it was demonstrated that the robot can navigate even when the maize plants are very small. Thus, the second objective was also met. However, weed detection was less than perfect. It turned out to be more sensitive to the light conditions than we had realized. Also, the turf patches were placed almost between the maize plants instead of well inside the row, and were out of the camera’s view. In conclusion, the robot is capable of a high degree of autonomy in the tasks of the Field Robot Event: it didn’t once get lost and it damaged few plants