Wageningen, The Netherlands
January 16, 2025
>> Winnaar vierde editie Autonomous Greenhouse Challenge bekend
Team IDEAS in the photo from left to right: Jury Leo Marcelis, Tao Lin, Wei Liu, and Jury Kathy Steppe.
The tomatoes have been harvested and weighed, the yields and costs calculated, and the results added up. With this, the winner of the 2024 Autonomous Greenhouse Challenge has been revealed. Team IDEAS emerged victorious with their algorithm for the autonomous cultivation of dwarf tomatoes, achieving the highest profit. Stef Maree and Silke Hemming from Wageningen University & Research (WUR) explain the choices that made the difference and the valuable insights this edition has provided for autonomous greenhouse horticulture.
The teams had been in suspense for months. After a summer of hard work developing algorithms for the autonomous control of their greenhouses, the cultivation phase began in early September. From that point on, their algorithms took full control of the greenhouse. It included strategies for the amount of lighting, heating, CO₂ levels, water supply, plant density, and the ideal harvest time. But did their decisions result in a good yield?
Cultivating dwarf tomatoes
This edition focused on dwarf tomatoes. Silke Hemming, project leader of the Autonomous Greenhouse Challenge, explains why this variety was chosen. “The fact that dwarf tomatoes only require a single harvest, can be picked by robots, and grow uniformly at the same height makes them ideal for autonomous cultivation. Additionally, their cultivation methods are representative of other potted plant cultivations. What makes growing dwarf tomatoes autonomously challenging is determining the right harvest moment. The plant needs to have enough ripe fruits, which can’t simply be inferred from the plant’s weight, as was the case with lettuce - the crop in the previous edition.”
The cultivation of dwarf tomatoes during the Autonomous Greenhouse Challenge.
Profit minus costs
Several factors were taken into account when determining the winner, explains Stef Maree, a data scientist involved in the Challenge. “The most important criterion was the net profit achieved by the teams. It was pre-determined that teams would earn money based on the number of ripe tomatoes per square metre of the greenhouse. This amount was then divided by the number of days it took to complete the harvest. The earlier the harvest, the better - provided the tomatoes were ripe, of course. On the other hand, the teams incurred costs for heating, electricity, CO₂, depreciation, and materials.”
Bonus and penalty points
In addition to the score for their harvest, teams could earn bonus points or incur penalties. Maree explains: “Bonus points were awarded for decisions related to Integrated Pest Management (IPM). Each week, teams had to decide how much and which types of biological pest control to use against potential threats, such as whiteflies. Correct IPM strategies earned them points. Teams incurred few penalties for manual interventions. This happened once when one team’s irrigation stopped, and in two cases, the algorithm miscalculated the harvest date, resulting in delays.”
Plant density made the difference
According to Maree, the winning factor was IDEAS’ decision to cultivate with as many pots per square metre as possible. “Normally, during cultivation, a grower spaces the plants further apart to ensure all leaves receive enough light and the plants maintain a good shape. But it turns out that this is not necessary for a good yield. The plant density of IDEAS was almost twice as high as that of most other teams, resulting in a higher profit per square metre per day. IDEAS’ overall cultivation strategy was also effective - they were resource-efficient, making extensive use of energy screens. The bonus and penalty points only caused some reshuffling between second and fourth place.”
Exceptionally high standards
Project leader Hemming reflects on a highly successful edition. “The teams operated at an incredibly high level. Apart from a few minor adjustments, each team managed to achieve a successful autonomous harvest. And they did so well before the 15 December deadline. This demonstrates that they developed excellent algorithms in a relatively short amount of time. As in previous years, the teams initially underestimated the amount of work and complexity involved. Especially during the testing phase before the challenge began, things often didn’t go as planned. But in the end, every team succeeded.”
A wide range of strategies
Hemming found the diversity of strategies both striking and fascinating. “Some teams went to extremes from day one. One team immediately implemented intensive lighting, while another did the complete opposite. One team barely heated their greenhouse at the start and then used a lot of heating towards the end. Such decisions generate highly interesting datasets. While high temperature or light levels can lead to a high yield, they also come with significant costs. There were also very different approaches to spacing out the pots. The harvest times were relatively close, with the first team harvesting after 70 days and the last after 80.”
The repotting of the plants used in the Challenge.
Advancing towards fully autonomous cultivation
According to Hemming, this fourth edition of the challenge has brought the horticultural sector closer to fully autonomous greenhouse cultivation. “Letting an algorithm take control of a greenhouse and achieving a full harvest after a few months doesn’t yet exist in practice. No grower has fully automated this process. However, specific aspects, such as autonomous temperature control, are already in use. We’ve demonstrated that cultivation - except for aspects like IPM - can be fully autonomous. Of course, there are still many challenges and areas for improvement, but we now have proof that it’s possible to complete a growing cycle with an algorithm.”
Flexibility to let go of standards
For Maree, this edition also showed that great results can be achieved with highly diverse strategies. “Of course, growers don’t always make the exact same choices, but the extremes seen in the challenge are far from typical. Yet, these choices still resulted in good harvests. This indicates that growers might have more room to let go of their standards and adapt their cultivation strategies to reduce costs - for example, when electricity prices are high. Once again, it’s clear that there are many paths to success.”
Herewith the full results of the Autonomous Greenhouse Challenge 2024:
The IDEAS team consists of six researchers from Zhejiang University in China, including a master's student, PhD student, postdoc and professor. Five of the team members have expertise in technology, one in horticulture.
Watch the video in which team IDEAS introduces themselves:
- Unfortunately, your cookie settings do not allow videos to be displayed. - check your settings
The Autonomous Greenhouse Challenge is supported by:
More information
Autonomous Greenhouse Challenge 4th Edition
We are doing it again! We are currently running the 4th edition of the Autonomous Greenhouse Challenge: the challenge for sustainable vegetable...
Artificial Intelligence in plant research
Discover how the experts at Wageningen Plant Research are using AI to advance sustainable agriculture.
Read more
Topics
Species
