Robots grow sugar beets after consulting the farmer

Harmen, a farmer, wakes up at 6 AM. He gets up, makes a cup of coffee and sits down at the kitchen table. His smartphone tells him that his autonomous tractors have done a lot of work that night. They have removed weeds and identified diseased plants. Everything has been recorded in high-definition photos. Harmen looks out of the window and asks the virtual assistant when rain is forecast. They discuss whether it would be wise to irrigate the crop today.

Later that morning, Harmen drives his tractor onto the fields. Sitting in the cabin, he tells his virtual assistant that the autonomous tractors can go to another field. The robots know exactly what they need to do there: spread fertilizer on the land in a very precise pattern, keeping exactly six metres away from the ditch. A drone takes off ahead of him to check where the birds’ nests are. If a robot has a malfunction or there is a problem in the field, the virtual assistant will alert Harmen immediately.

Harmen’s hi-tech farm is not yet a reality, but that day is coming ever closer. More and more robots are being used in agriculture and they are becoming more advanced. The global market for farm robots is predicted to increase by a factor of three in the next few years, from 13.5 billion dollars in 2023 to 40.1 billion in 2028. ‘Current technology is already really smart,’ says Paul van Zoggel, the precision agriculture programme manager at WUR. ‘Robots can do a lot: sow seed, remove weeds, mow the grass and so on. All arable sectors are relying increasingly on sophisticated robots.’

Using robots is a necessary measure for many farmers because of the sector’s ageing population profile. Only 10 per cent of Dutch farmers are under 40, according to data from Statistics Netherlands (CBS). This percentage has barely changed in recent years, whereas the proportion of farmers aged 67 and over has increased to more than 20 per cent. It is also becoming increasingly difficult to find farm labourers willing to work whatever the weather. ‘Fewer skilled workers are coming from other countries,’ says Van Zoggel. ‘That’s because of the strong economic growth in Eastern Europe and because you can earn more money in the Netherlands in jobs that are less physically demanding.’

Robots don’t just fill those vacant positions, they also have other advantages. They don’t need breaks and they can often work 24 hours a day. They are more precise, reducing the amounts of fertilizer and crop protection products that are required. Furthermore, many robots are lighter and smaller than conventional agricultural vehicles, causing fewer problems with compacted soil as a result.

But the rapid progress in the robotization of agriculture is also bringing new challenges. ‘The bottleneck is the problem of how to manage all the technological equipment,’ says Van Zoggel. ‘From a technical point of view, the robots are complex and efficient, and you can achieve various environmental targets because the computer systems are designed to focus on that. But they aren’t easy for farmers to work with. The farmer basically has to have an affinity with technology and data and software.’

Some farmers do have that affinity, says Van Zoggel: ‘I know one who has a kind of ground control room, like an air traffic control centre. In America, a lot of farmers have been using automated tractors for a while. I’ve heard they look for youngsters who play the video game Farming Simulator because those gamers are better able to operate the machines.’

Van Zoggel wants all farmers to soon be easily able to manage their robots and even hold complete conversations with them in the farmer’s natural language. The ultimate aim is to have an integrated interface that combines all the systems the farmer uses on their farm. This should be a platform where the systems communicate with one another and with the farmer.

What Van Zoggel has in mind is something similar to the superhero Iron Man’s digital sidekick J.A.R.V.I.S. Iron Man talks to the all-knowing assistant in his helmet, and J.A.R.V.I.S. then controls the weapon systems while giving the superhero updates and warnings. ‘Some farmers do have a digital assistant, but it’s still not easy to discuss things with it. We eventually want to get to a situation where a farmer can instruct their virtual assistant to spread fertilizer, and the assistant can say in response, “Are you sure? Because it’s about to rain.” The system will be able to help the farmer come up with the right solutions because it knows the local conditions and the applicable regulations.’

Ard Nieuwenhuizen is one of the researchers working on this integrated interface at WUR. He spent years in the private sector working on the development of autonomous farm vehicles, and moved to Wageningen in 2018 to work in the Vision & Robotics programme.

Nieuwenhuizen has also noticed how farmers have to put a lot of time and effort into managing their machines. ‘Farmers often have to go to their computer and log into a web application. Each robot is a different brand and uses a different application. And those robots have difficulty communicating with one another when they are in the same field.’

‘There is a real need for better interaction between the farmer and the various systems,’ he continues. ‘Farmers should be able to talk to their robots in the same way they normally talk to their workers.’ This might have sounded like science fiction 30 years ago, but today, in 2025, millions of consumers ask Alexa and Google Assistant to turn the lights on or order some washing powder. The options are increasing thanks to the rise of ChatGPT and similar large language models (LLMs).

These LLMs will be used more and more in farming too. Nieuwenhuizen cites the example of the chatbot Botato. ‘It has absorbed vast amounts of information about the cultivation of potatoes. Now, farmers can ask the chatbot all kinds of questions, for example: “Can you predict my potato harvest? What about if I harvest one week later?”’

To create Botato, the developers at Wageningen Plant Research had to feed the app with crop growth models that calculate potato yields. For a complete, integrated interface, the Wageningen researchers will need to collect many terabytes of information that the AI system can learn from. Otherwise the system will start hallucinating, says Paul van Zoggel. ‘Then the model will make up answers. We need to make sure the model bases its answers on scientific knowledge.’

There is plenty of such scientific knowledge available in Wageningen. Van Zoggel explains, ‘We’d like to feed the AI models with all the knowledge we have in-house on the agronomy, economics, ecology and social science aspects of agriculture. That knowledge is currently quite fragmented, so we need to bring it all together.’

While Van Zoggel works on collating Wageningen’s knowledge about agriculture, Nieuwenhuizen is focusing on the technical side of the interface. ‘The robots need to become even more intelligent: the devices need to record information and make it available digitally. For example, a weeding robot will need cameras to see what goes into the machine and what comes out of it. A camera at the front would record the crops and weeds. The robot would then remove the weeds and a camera at the back would check that the plants have not been damaged.’

Farmers will also need to modify their farming methods to accommodate these new machines. ‘Cultivation systems will need to change in line with the technology. To give an example, we are already seeing changes in apple tree shapes to let the robots reach the apples. The idea of tree shapes being changed is not new — fruit growers have for some time had bush trees with short trunks and espalier fruit trees. But nowadays, robotization is increasingly a factor driving the changes. The shape of apple trees is chosen to make it easier for robots to manoeuvre, see the fruits and pick them.’

A lot of work is needed before farmers will be able to discuss the weather, potato harvests and nitrogen emissions with their robots. Nieuwenhuizen expects that kind of communication to be possible within five years. ‘Language models like ChatGPT are developing incredibly rapidly and chatbots are constantly improving in quality. If farmers can chat with these systems in the future and get decent answers, that would be a major advance for agriculture.’