A Robotic System for Plant-Scale Husbandry

Abstract The object of the work was to demonstrate individual spot treatment with spray of transplanted cauliflower from a small vehicle that could operate autonomously within a field. The only prior information required was to be the plant spacing, an approximate row length and the number of beds. To achieve this a small horticultural toolframe tractor was fitted with navigation and computer control equipment and an experimental marking device. A charged couple device (CCD) was located centrally at the front of the vehicle looking forward and down in order to provide images of crop rows. The images were analysed to provide guidance information. Signals from vehicle motion sensors (odometry and accelerometers) and a compass were combined with the image analysis data to provide vehicle control information that was robust to natural variability. A segmentation was performed using image intensity, size and geometric information to pick out crop plants requiring treatment. This information was passed to the treatment device controller which selected from an array of 27 solenoid valves to operate appropriate spray nozzles arranged transversely across the machine. Field trials of selective spray treatment of a young transplanted crop, conducted at a vehicle speed of 0·7 m/s, showed that all plants were treated, though coverage of each plant was not perfect. Mean measured error along the row was 1 mm and lateral error 11 mm. The standard deviation in these errors were 26 and 20 mm, respectively. Only 8% of the total area of the bed was covered with spray, thus showing a potentially large saving in agro-chemicals with environmental and economic advantages. The vehicle has been operated extensively in a wide variety of crops of different sizes and ground coverage, with good reliability under favourable lighting conditions.