Greenhouse competition: teams spend summer building algorithm that will control greenhouse autonomously

Growing crops in a greenhouse without human intervention. It sounds like a utopia, but according to Stef Maree, data research associate at WUR Green Horticulture, it’s not such an unrealistic prospect. “Greenhouses are getting increasingly 'smarter'. Different technologies already ensure that water and nutrients are administered at the right time and that an optimal temperature and humidity are maintained. Data collection also increasingly enables growers to make better decisions. This results in more efficient cultivation of crops. Which is very important, given the challenges we are facing such as increasing food demand, higher energy prices, carbon emissions reduction and labour shortages.”

WUR forerunner in autonomous cultivation

According to Maree, WUR is one of the world leaders in autonomous growing. “Within WUR, a large team of researchers is involved in various aspects of autonomous greenhouse growing, from design to operations. One of the big projects we’ve worked on in recent years is AGROS. In this project, we showed that we can control cucumber cultivation completely autonomously. Growing based on AI models and expert-based models produced results similar to when growers controlled the greenhouse themselves. Proving the practical feasibility is a huge breakthrough in autonomous growing.”

Autonomous Greenhouse Challenge

Besides through its own research, WUR is also trying to contribute to the development of self-operating greenhouses in other ways. For example, through the Autonomous Greenhouse Challenge. Maree explains what this competition aims to achieve. “Through the Challenge, we want to bring together people with different areas of expertise from around the world. From plant experts to technology experts. It also offers participants a unique opportunity to gain hands-on experience with autonomous growing. Testing technology in greenhouses is expensive. By making our research greenhouses available, students, researchers and companies can test their tools for free and in a fun way.”

Growing dwarf tomatoes

This year will be the fourth Autonomous Greenhouse Challenge. Every ear, a different crop takes centre stage, says Maree. “This year we chose the dwarf tomato. Unlike the plants in previous challenges, this one is in pots and is sold as a whole plant. That in turn requires new choices, because there is no single autonomous system for every crop or greenhouse. The challenge is to sell the plant with both ripe and almost-ripe tomatoes, so consumers can watch the last tomatoes turn red at home. We try to make the Challenge a little more elaborate and difficult every year. For example, by creating less and less room for manual adjustment. We do that because greenhouses are also becoming increasingly more autonomous in practice.”

First part completed

A total of 23 teams from all over the world are taking part in the Challenge, the first part of which was completed in the spring. The Challenge required the teams to do different tasks. Maree: “The first task was to develop an algorithm or model that they had to use to estimate the status of a dwarf tomato plant based on a photo. They also had to grow dwarf tomatoes virtually for two months. The final phase took place on 6 and 7 June in Bleiswijk, where our research greenhouses are located. Here, the teams had to develop an algorithm for identifying whitefly on yellow sticky traps in a hackathon. These flies can affect plant growth. It’s therefore important for growers to keep this pest under control.”

Five teams compete for the win

Part 1 of the Challenge eventually produced five teams who will compete for the win. These five teams have to develop an algorithm that can control a greenhouse completely autonomously, Maree says. “Each team is given its own greenhouse in which they will have to grow and eventually harvest dwarf tomatoes. In their algorithm, they can turn various buttons, such as water release, temperature, light and humidity. The trick is to find an optimum in this. More light and heat means you can harvest earlier, but also that more energy is consumed. The greenhouses are equipped with standard sensors, but the teams may fit additional sensors.”

Developing the algorithm

Throughout the summer, the teams are working on developing their algorithm. To do so, they are using a WUR simulator. Maree: “This simulator is a digital version of the tomato they will be growing. We provided the simulator with data from a previous trial. During that trial, we grew dwarf tomatoes in the same greenhouse, but manually. That means we have a better idea which parameters are important for the crop to develop and which sensors are needed for that. The teams will build their algorithm entirely using this simulator. In practice, they then just need to fine-tune whether the greenhouse responds as they expect based on the simulation.”

Most lucrative crop wins

The algorithm must be ready on 2 September, says Maree: “After that, the greenhouse will need to do the rest of the work. Tomatoes must be harvested by 16 December. The winner is the team with the highest net profit. That depends on the price per pot, the number of pots worth selling and the costs incurred. We pre-set the price per pot at 1.80 euro. Fewer than average red fruits will mean the price is lower, more red fruits means the price is higher. Flavour also influences the price. We base this on the dry matter content. The higher this content, the more flavourful the tomato should be.”

Valuable knowledge and insights

Apart from eternal fame, the winner also gains valuable knowledge from the Challenge, says Maree. “They can use their algorithm for further research or apply it operationally. Several successful businesses have emerged from previous editions. For us too, the Challenge will bring new insights. For example, whether new sensors provide reliable data. But also which strategies will be most efficient. Because ultimately, that’s the goal of autonomous farming. As an organiser, we don’t see exactly how the algorithms work, but we do see the consequences of choices. That in turn provides us with loads of valuable data, which we will be able to use in further research and to help the industry move towards fully autonomous greenhouses.”

The 4th Autonomous Greenhouse Challenge is organised by Wageningen University & Research, BU Horticulture and Flower Bulbs and sponsored by Tencent, Biobest, Fluence, Lensli Substrates, Pöppelmann, Quantified, Certhon, Vreugdenhil, Gebr. Geers bv and LetsGrow.