Most fruits you see today require a process called pollination to bear fruit. Pollination is when pollen released from the stamen adheres to the pistil. This reproductive activity produces tasty red strawberries.
In the case of strawberries, stamens and pistils exist in the same flower and whether the pollen from the stamen reaches the pistil determines whether it bears fruit.
In nature and on farms, pollination occurs when bees and other insects move over flowers to collect nectar.
In pollination, it's essential to distribute pollen evenly on the pistil. Otherwise, the strawberry fruit will not grow in an even shape.
Plant factories are very harsh environments for native insects used for pollination.
Strawberry flowers are not very nutritious nectar-producing plants, and no other plants grow in the plant. So bees cannot survive sustainably in such an environment. Therefore, plant factories usually remove the bees and reintroduce new hives every few weeks or months. However, honeybee populations are declining globally, and there are sustainability and ethical issues with disposing of living creatures.
The bee installation itself is also one of the risks of disease outbreaks in plant factories. Dead bees can be a source of mould, and there is a risk of diseases from new hives.
On the other hand, bees cannot control how many times they pollinate a single plant or how evenly they deposit pollen on the pistils. As a result, current farming methods inevitably result in unstable fruit appearance, taste and quantity. Our robotic pollination and management system can improve efficiency and quality.
We are developing technology that mimics the pollination behaviour of bees, which have coexisted with plants for a long time.
Our technology can indirectly extract features of the pollination behaviour of honeybees by using neural networks to estimate their skeletons from videos of bees collecting nectar and pollen.
We are testing to apply the extracted features to pollination algorithms to form more beautifully shaped strawberries.
To distribute pollen evenly on the pistil, the direction of contact of the pollination brush must match the order in which the flower is facing.
Conventional detection methods can only identify the position of the flowers. They cannot recognize the direction of flowers and from where the pollination brush should approach.
HarvestX has developed a machine learning technique that uses 3D models and rendered images to generate supervised data on flower orientation. By applying a neural network that detects the direction of a person's face, we have established a technique for estimating the directions of flowers.
This technology will bring stable production of beautiful strawberries to plant factories, which honeybees can't achieve.
HarvestX is developing a highly accurate yield prediction system using flower and fruit maturity information. The system statistically processes data on the flowering status of each seedling and the growth rate of the fruit to create a more productive growing environment for seedlings.
HarvestX developed a fruit maturity classification algorithm before the yield prediction system. This algorithm classifies strawberries into four detailed stages, from the initial fruiting state to ready for harvest. Using the algorithm to determine the fruit's maturity makes it possible to estimate the harvest time, preventing human judgment variations correctly.
In developing agricultural robots, the number of trials is limited because they deal with natural things, which is a unique challenge of this field. For example, it is impossible to repeat harvesting experiments multiple times on the same strawberry fruit. For this reason, it is essential to develop robot systems based on simulators.
As we have our plant factory, we have been developing simulators that reflect the challenges of the actual environment and significantly improve the efficiency of the development of robotic systems.