Autonomous pollination of individual kiwifruit flowers: Toward a robotic kiwifruit pollinator

Abstract

There is an increasing concern that the traditional approach of natural kiwifruit pollination by bees may not be sustainable. The alternatives are currently too costly for most growers due to high labor requirements or inefficient usage of expensive pollen. This paper presents a performance evaluation of a novel kiwifruit pollinating robot designed to provide a more efficient, reliable, and cost‐effective means of producing kiwifruit. The robot comprises a novel air‐assisted sprayer, a machine vision system employing convolution neural networks, and a flower targeting system for efficient and effective application of pollen to individual flowers. We show that this pollination system is capable of individually targeting and pollinating 79.5% of flowers at 3.5 km/hr while using comparable amounts of pollen to commercial Cambrian operators. Furthermore, flowers that were successfully pollinated at 1 km/hr grew into the first robotically pollinated kiwifruit which were comparable in quality to commercially grown kiwifruit. However, the overall fruit set was found to be well below commercial requirements and further work on increasing the overall yield is required.

A high‐resolution, multimodal data set for agricultural robotics: A Ladybird's‐eye view of Brassica

“…Many architectures have been proposed to solve this task including Faster R-CNN (Ren et al, 2015), SSD (Liu et al, 2016), R-FCN (Dai, Li, He, & Sun, 2016), and YOLO (Redmon, Divvala, Girshick, & Farhadi, 2016). These methods have been used successfully in agricultural applications to recognize and locate fruit (Bargoti & Underwood, 2017;Sa et al, 2016;Williams et al, 2019). While these methods work well in cluttered scenes where fruit is likely to overlap and be partially occluded, fruit generally produces a distinct visual signature.…”

“…Systems that approach this level of autonomy are rare and typically address well-defined problems at the frontiers of research. Examples include robotic harvesting (Bac, vanHenten, Hemming, & Edan, 2014), pruning (Botterill et al, 2017), pollination (Williams et al, 2019), and weeding (McCool et al, 2018). To develop these systems and make them economically viable, further research and development is required.…”

Improvements to and large‐scale evaluation of a robotic kiwifruit harvester

“…Similar to targeted weed‐spraying robots, autonomous pollinators are being developed to target and spray crops directly to improve the efficiency of growing quality fruit (Amador & Hu, ; Berman, Kumar, & Nagpal, ; Kurosaki, Ohmori, & Takaichi, ; Maghsoudi, Minaei, Ghobadian, & Masoudi, ). A kiwifruit pollinator has been developed in conjunction with the presented kiwifruit harvester, and has been shown to be capable of individually pollinating 80% of flowers in an orchard and producing kiwifruit of commercial quality (Duke et al, ; H. Williams et al, ).…”

“…A kiwifruit detection system using semantic segmentation was able to detect 76.0% of kiwifruit in a real‐world orchard (H. A. Williams et al, ). Furthermore, a kiwifruit flower detection system using Faster‐RCNN was capable of detecting 79.8% of kiwifruit flowers, indicating its potential use for detecting kiwifruit under the same conditions (H. Williams et al, ).…”

Flower Interaction Subsystem for a Precision Pollination Robot

“…The idea of using robots to aid pollination has been considered for more than a decade [14]; however, research in this area is quite limited beyond conceptual designs [15]- [17], only a few systems have been demonstrated in practice [18]- [20], and even fewer with autonomy [8], [9]. The systems developed in [8], [9] use sprayers for pollinating flowers of tomato and kiwifruits, respectively instead of physically touching each flower like bees would do.…”