Using data to work smarter together: AI companies in Brabant are working towards a breakthrough in agricultural robotics
How can we ensure that AI systems become smarter without every company having to reinvent the wheel? That is the question Brabant.ai – a programme run by Brainport Development in collaboration with AI companies in Brabant – is working to answer.
The project focuses on robotics in agriculture, and more specifically on robots that can identify and remove weeds. Various companies in Brabant are building such weed-removal robots, such as Odd.Bot, Trabotyx and Pixelfarming Robotics. But there is a challenge: AI models that recognise plants or weeds require a great deal of image data to learn effectively. And that data often varies significantly depending on the robot, camera, season or lighting conditions.
The result?
Every company has to train its AI model largely on its own. That takes time and money.
Within the project, AI companies in Brabant are therefore investigating whether shared data can help to make AI models faster and more effective. In a so-called proof of concept, they are testing whether images from different robots and cameras can be converted into a single usable dataset using AI. If this proves successful, recognition models can become more robust and less dependent on specific hardware.
Weed-control robots make it possible to remove weeds with great precision and in a targeted manner. Instead of spraying an entire field with chemical plant protection products, a robot can intervene only where it is really necessary. This has several advantages: farmers use fewer chemicals, the soil and biodiversity are less impacted, and water quality improves because fewer chemicals end up in groundwater and surface water. At the same time, automation helps farmers work more efficiently and address the labour shortage in agriculture.
But the project is about more than just technology.
A second key objective is to learn how companies can share data with one another without compromising their competitive position. In practice, this often proves to be a major barrier to AI development. That is why around ten AI companies in Brabant, with expertise in areas such as computer vision, data governance and robotics, are collaborating on a practical approach to data sharing.
This approach will ultimately consist of a toolkit and practical guidelines: how do you agree on data ownership, how do you build trust between parties, and which collaboration models work in practice?
The insights from the agricultural project form the basis for this. The aim is for the approach to eventually be usable by the entire Brabant AI community, which now comprises around 180 companies.
Brainport Development acts as project leader within the project through the Brabant.AI programme and facilitates connections between businesses, knowledge institutions and the government. Breda Robotics supports the project as co-project leader and contributes expertise in the fields of robotics and agile innovation.
In this way, Brabant is once again demonstrating how collaboration can help to accelerate innovation, not only in agriculture, but potentially also in many other sectors where AI and data play a key role.
Teaching a weed-control robot to recognise weeds is trickier than it seems. This is because plants can look very different depending on the season, the stage of growth, the type of crop and the lighting conditions. Cameras and robots also differ from one another, meaning that images are not always consistent. What is clearly a weed in one field may look very similar to a young crop in another. AI systems therefore require a large and varied set of image data to reliably learn to distinguish between crops and weeds.