Pesticide spraying, a particularly heavy burden on farmers
Reguminwas founded in 2018 by Takuhiro Naruse and Yoshihiro Noge with a desire to preserve Japanese food culture. When they first started the company, they used their knowledge of the komatsuna cultivation business they conducted in Shizuoka Prefecture as the basis for a project in Fukaya City, Saitama Prefecture, calledDEEP VALLEY Agritech Award 2020In.Proposed robotic pesticide spraying serviceThe first time we were awarded the top prize, we started our activities in the city.
Regumin's goal was to develop a pesticide spraying robot,Among agricultural work, pesticide application is particularly burdensome.This is because of the need to prevent pests and diseases from spreading. Pesticides include not only insecticides, which are sprayed when pests appear, but also fungicides and fungicides, which are applied preliminarily, amounting to 20 to 25 different types of pesticides throughout the year.
In Japan, where most farmland is cultivated on a small scale, it is difficult to install large machinery and there are concerns about noise pollution in the neighborhood, so spraying is done mainly by human labor, which imposes heavy labor costs. In addition, there are concerns about hose routing, wearing hot protective clothing, and inhalation of pesticides,The physical and psychological burden on farmers is also great, leaving room for automation by robots.Mr. Narusei thought.
However, unlike warehouses or agricultural greenhouses that are moved under certain conditions, the ground in open-air cultivation is dirt and can be bumpy or slippery in places,Difficulty in controlling autonomous driving has made it difficult to commercialize agricultural robots that actually work outside.It has been considered to be a
GPS is a convenient way to obtain location information for autonomous driving, but even with high accuracy, there are errors due to satellite positioning and radio interference,Correction by image recognition as well as GPS is essentialMr. Narusei felt that the LiDAR system was not suitable for his needs. At first, we tried to use another manufacturer's CPU board, but the processing performance was not enough because of the large amount of point cloud data to be acquired by LiDAR," Narusei says.
Ideas made possible only with Xavier and patents obtained
At such a time of year.NVIDIAMr. Narusei attended GTC Japan, an AI conference hosted by the Jetson AGX Xavier, and saw a demonstration of a robot equipped with a Jetson AGX Xavier running autonomously through an office.Xavier can enable robots that contribute to actual agricultural work.I had a gut feeling that it was a good idea, and I bought Xavier on the spot.
Xavier is a single-board computing platform that can handle OpenCV computations on the edge, as well as the CUDA processing required for deep learning. Its ease of use made it the only choice." said Narusei.
In addition, to gather information, the NVIDIA Inception Program, NVIDIA's startup support program (https://www.nvidia.com/ja-jp/startups/), which also deepened our knowledge of machine learning and image recognition.
Regumin's pesticide application robots are equipped with numerous sensors, including cameras, magnetometers, GPS, and temperature sensors, but Xavier'sBrain to process information obtained from all sensorsIt functions as a
Using Ubuntu onboard Xavier, the robot recognizes the terrain of the cultivated land using laser sensors in addition to GPS, and operates with programs written in C++ and Python. The 3D point cloud data is processed by Xavier to obtain coordinates, which are used to recognize the shape of the ridges* and to align the wheels with the hollows. This method enables highly accurate autonomous driving with a margin of error of 1 to 2 cm, for which a patent has been obtained.

By using Regumin's pesticide spraying robot,The time required to spray an area per hectare is reduced to 250 minutes, compared to 400 minutes per hectare when a person works with a power sprayer.This year, the company has been able to achieve its goal of 150 minutes of actual operation time for the pesticide application service. Furthermore, this year, by operating multiple robots simultaneously, the actual operating time of the pesticide application service is targeted to be approximately 150 minutes,Time savings of 60% or moreThe company has set forth the following The cost per hectare of work is also reduced compared to the cost of work done by a person using a power sprayer.45% cost reductionThe goal is to achieve the following
Additional features will be considered for inclusion in the future.
Regumin is also conducting research on early detection of disease and insect damage, including at the Saitama Institute of Technology,NVIDIA DGX SystemownCollaboration with universities to conduct joint research using machine learningWe are also working on
With up to 32 TOPS of computing power, the Jetson AGX Xavier module has enough power to incorporate these research findings into new robot functions in the future, which is one of the reasons Narusei chose Xavier.
Plans to expand the scope of services and areas of installation, with an eye toward global expansion.
Pesticide spraying robot is able to operate according to the width of the ridges.Therefore, it can be applied to other crops. Regumin is already planning to contract out its new pesticide application services to broccoli and cabbage, specialty crops of Fukaya City.
In addition, the company is looking to expand its services not only to other prefectures in Japan, but also to the global market in the future. In Southeast Asia in particular, there are many cases where crops are grown on small plots of land, as is the case in Japan, and Mr. Narusei expects that the services in Japan can be applied to this area.
Mr. Narusei stated the following. "The proportion of human workload in agriculture is large,Adoption of AI and other technologies is a mustIt is important to distinguish between tasks that are performed by humans and those that are automated by machines. It is important to separate the tasks that are performed by humans from those that can be automated by machines. Human knowledge and experience are required in areas such as designing fertilizers according to weather conditions and determining the varieties to be grown. Also, in areas such as research and development, sales channel development, and marketing, the quality of the product will be higher if it is done by humans.Rather than replacing everything with a machine, work can be subdivided and roles assigned based on an understanding of the user's needs, thereby increasing efficiency and improving quality.We think it's a good idea."
Ridges: Long, narrow, linear patches of soil in a field where crops are planted or sown.









