AVO in agronomic test at KWS

KWS is one of the world’s leading plant breeding companies. It has more than 5,500 employees in 70 countries. For 160 years, KWS has been independently and autonomously managed as a family business. Its main focus is breeding plants and producing and selling maize, sugar beet, cereals, canola, and sunflower and vegetable seeds. 

Under the project “Future Live – Robotic Weeding in the field, KWS organised in collaboration with the University of Göttingen and the Sugar Beet Research institute (IFZ) an agronomic test of the robotic systems used for weed control in sugar beet. So we went with AVO to KWS, near Northeim in Germany, 1000 km from our workshop.

The aim of this project was to analyze how well each robot could get rid of the weeds and to evaluate the on-board technology.

While waiting for the results of the analyses, this is an opportunity for us to tell you in a little more detail what’s under AVO’s hood and how it all works!


Machine Learning 

Machine learning is a form of artificial intelligence that allows computers to learn how to do things without being explicitly told how to do them. This technology makes it possible to develop computer programs that can adapt to new data.

An algorithm, without which nothing could work, is at the heart of the system. By analogy with the human brain, it simulates a neural network.

In practice

The robot has cameras, which make the flashes you may have seen in one of our videos

It takes several hundred photos per square metre.

For each picture, the computer first determines whether there is a plant in the image. If there is, it works out whether it is a crop or a weed. Once a weed has been identified, the robot scans its position and transmits that information to the spraying module at the back. The computer then chooses the best nozzle corresponding to the position of the weed, which sprays the target with a micro-dose of herbicide.


The algorithm feeds on pilot projects

The essence of this technology lies in feeding the database with information. We work with nature but the huge diversity of colours and shapes makes it very difficult for a computer to identify objects in images. There are many confusing variables, such as different weather and lighting conditions, but also the diversity of plants and their different growth stages, as well as the topography of the land and the different things that can appear in a photo. 

The more times the algorithm is tested on pilot projects in a variety of conditions, the more powerful it becomes. Although many conditions can be replicated in our greenhouses, each pilot is a phenomenal accelerator of this technology, which didn’t exist just a few years ago.


Photo 1 ©SpiekerFotografie / Video : by KWS