This year will see the start of the 3rd edition of the ‘Autonomous Greenhouse Challenge’. The challenge is to grow lettuce in a greenhouse ‘completely autonomously and without human intervention’ through the use of AI. The lettuce also needs to meet a high standard of quality and yield. It also has use as few resources as possible, such as water, fertilizer and energy. The participants had to compete against a grower who is allowed to walk around their own greenhouse and check up on their crops.
Cucumber (2018) and cherry tomatoes (2019/20) were grown autonomously in previous editions. The teams had to determine the ideal for temperature levels, amount of light, irrigation and a number of other parameters, such as the density of plants and their stems.
The teams used a number of standard sensors in the greenhouse for their crops. They were also allowed to install their own sensors and cameras in the greenhouse to collect additional information while the crops were growing.
Can technology beat growers?
In the first two editions, teams took part with employees from tech giants such as Microsoft, Intel, Tencent, NXP and Samsung. The first competition was won by a team made up of Microsoft employees and students from a Danish and a Dutch university. The second event was won by a team consisting of employees, students and researchers from Hoogendoorn Growth Management, Van der Hoeven Horticultural Projects, Delft University of Technology and Keygene. This team performed better than the reference team of growers across all sections of the competition. Which led to the question whether ‘technology can already beat growers and does this make people redundant?
We have been using all kinds of technology in greenhouse crops for many years. Growers have long since dispensed with having to open the windows themselves when it’s too hot or fire up the furnaces when it’s cold outside. For more than 50 years, climate computers have been in existence that have taken over that job from them. These kinds of adjustments are made automatically by linking temperature readings to a controller that can open the windows or turn up the heating. Knowledge systems were already being utilized in horticulture 30 years ago with the aim of making things run even smarter. These systems, which contain rules incorporating ‘human knowledge’, are now seen as the forerunners of AI.
Basic climate control system
In a standard, basic climate control system, for example, you would turn the heat up in a greenhouse if the thermometer registered a temperature that was 1 degree too low. However, if a gardener was still regulating that manually, then they would probably not make that choice if they knew that the next day was going to be very hot. That’s because they know that the average 24-hour temperature was much more important to the plants than a transitory temperature.
In other words, they might let it drop one degree colder now because tomorrow, when it will be very warm, the temperature difference would then be made up for ‘free’ because the greenhouse will be heated by the sun. This simple example illustrates how the gardener regulates the climate in their greenhouse based on what they know, on their experiences and their gut feeling.
So, it seems logical to make sure that all that knowledge about growing crops is gleaned from growers’ heads and fed into a model. Surely this would then lead to winning the next autonomous greenhouse challenge -and without any help from human hands either!
Digitalize the green thumbs
However, doing that without the physical help of people is not yet possible for the time being…. First of all, the participants in the challenge only control a limited number of parameters during the growing process. Besides, a lot of manual work still needs to be done by people inside a greenhouse. All kinds of tasks, such as trimming leaves, checking to spot diseases and pests in time, and harvesting fruit and vegetables, to name a few things, can still not be automated. Although they are currently working very hard on changing that too.
More importantly, AI models that are used rely on people as well. Consider AI specialists, but definitely also those people who possess horticultural knowledge. Not everything can be simply codified or measured. A grower can ‘read’ their crop extremely well without always being able to formulate this in explicit rules or words. Moreover, this concerns a living crop and a highly variable environment. Much work still needs to be done in order to digitalize the green thumbs of the gardener.
That said, the first two editions of the challenge do suggest that a lot is already achievable. Especially if AI and technical specialists work together with people who have that expert knowledge of growing crops. What we hope is that it will be possible over the shorter term to use models that are capable of making everyday decisions. This leaves more time for those who do have that horticultural knowledge to spend on solving more problematic issues and special cases. In short, making better use of the fewer people available who know how to grow crops. Hopefully making more use of technology as well!
Robot trainer and data specialist
So no, people are by no means redundant. Not now, nor in the foreseeable future either. However, we do need other kinds of people. Not a tomato picker, but rather a robot trainer, data specialist and people who create the tools to digitalize green know-how.
If you are a technical specialist or student who would like to have a look around the horticulture world, then there are plenty of opportunities in our beautiful sector. Who knows, as an AI specialist, you might even want to try your luck in the third edition of the Autonomous Greenhouse Challenge. Send me a message, and perhaps I can hook you up with lettuce growers who would love to team up with you for that challenge!
For info on the 2021/22 challenge, please visit this website.
Photo: GearSense (from Gearbox Innovations) used for data analysis in the second challenge by the winning team. Photo amongst a tomato crop.
This post was first published on InnovationOrigins