These 5 teams are to compete in 4th edition of the Autonomous Greenhouse Challenge

The hackathon event marked the end of the first phase of the Autonomous Greenhouse Challenge. Twenty-three international teams from around the globe entered the Challenge, with only the top five teams advancing to the next phase of the competition. These five teams emerged from the online challenge completed by each of the 23 teams on 31 May.

During the online challenge, the teams had to detect dwarf tomato plants using computer vision and grow virtual dwarf tomatoes in a simulated greenhouse environment provided by WUR. The teams were then challenged during a five-hour hybrid (onsite and online) hackathon event on 6 and 7 June to complete an additional task and pitch their approach in front of a jury. After the hackathon, the five finalists were announced.

Enthusiastic teams and jury

Stef Maree, a machine learning researcher at WUR, took to the stage first. Maree: “A total of 200 team members competed online, all eager to secure one of the five greenhouse compartments for this year’s challenge. We saw that all teams performed very well during the computer vision tasks and simulator tasks over the past couple of months. And yesterday I also saw a very enthusiastic jury judging all the pitches. As a bonus task, we gave the teams a crash course on diseases and pests, such as catching insects on sticky traps and automatically detecting and analyzing beneficial and harmful insects. It was really impressive what all the teams built with just four training images. During such a hackathon, every detail has to be correct. The challenges proved hard to tackle, but some teams excelled!”

The five finalists

A very enthusiastic head of the jury, Kathy Steppe from Ghent University, concluded the hackathon results and announced the moment everyone had been anxiously waiting for: the five finalists of the 2024 edition of the Autonomous Greenhouse Challenge.

These five teams will get their own greenhouse compartment at the WUR facilities in Bleiswijk to remotely and autonomously grow a dwarf tomato crop assisted by the technology provided by WUR and the sponsors of the event. The five finalists, according to the scores they delivered during the online challenge and the hackathon, are:

1. IDEAS - Zhejiang University, China
2. MuGrow – TU Delft, Gardin, Rijk Zwaan, Wageningen University
3. AgriFusion – Croft, IMEC, GreenBites, Harvard University, Korea University of Technology and Education, Seoul National University
4. Team Trigger – Grit, Ridder, Daeyoung, Bigwave, Seoul National University, Keimyung University
5. Tomatonuts - Wageningen University, China Agricultural University, Jingwa Agricultural Science and Technology Innovation Centre, Golden Scorpion

Next steps

As part of the next steps of the challenge, the finalists must first provide a list of sensors required for their strategy to grow dwarf tomatoes. Before 15 July, the teams need to share the first version of their control scripts with the jury. By 1 August, all sensors must be available at WUR Bleiswijk for installation in the respective greenhouse compartments. Then, the five teams have to provide the final version of their control scripts, and after 23 August, no changes in the control scripts and algorithms are possible anymore.

From 2 September to 15 December, the teams will put their machine learning and computer vision skills to the test by remotely and autonomously growing a real dwarf tomato crop in their own greenhouse compartment. Their goal will be to achieve the highest yields and best quality with the most sustainable input of resources, such as water and energy, and thus maximize net profit. New is also that they have to make decisions on the biological pest control actions. On 16 January 2025, the winning team of the 4th Autonomous Greenhouse Challenge will be announced by the jury during the final event.

Three years of challenges versus research

During the hackathon event, moderator Silke Hemming, head of the Greenhouse Technology research team at WUR, asked the jury to share the learnings of the past three Autonomous Greenhouse Challenges and how to implement these in their respective research domains. Kathy Steppe noted an evolution in how the teams controlled their greenhouse compartments over the past three challenges. “I believe that relates to the developments we see in artificial intelligence (AI). I see the power of combining mechanistic models with AI models to find out how to go from an instant observation towards long-term growth of the plant.”

Jury member Leo Marcelis from WUR observed that especially during the first Autonomous Greenhouse Challenge, AI and vision weren’t yet used to the current extent. “I had expected more development in the use of available sensors and I expect more sensors to be used during the upcoming challenge. Sensors get cheaper so become more attractive, but how to make use of available sensors, that remains to be the challenge from my perspective. In the end, AI and vision have to compensate for the upcoming lack of labor and lack of experience and knowhow of greenhouse growers in the future.”

In-Bok Lee, jury member from Seoul National University, emphasized the potential of cross-innovation between different high-tech sectors. “We can use the technology developed in different other sectors to propel innovation in greenhouse growing. But we have to keep focusing on centralizing the grower.”

Need to redesign crops

Jaap Weerheim from one of the company sponsors, Certhon, went on to say that “if we want truly autonomous greenhouses, we need to dare to redesign how and what crops we grow. We need easy crops and that requires new visions on crop breeding. Can a crop be grown autonomously? This approach will surely require time to land in everyone’s mindset.” When asked about the differences in greenhouse growing in the Netherlands compared to abroad, Weerheim noted that the Dutch already have a lot of experience in horticulture. “That can actually hinder growers and manufacturers from embracing new technology. I think we have to rethink automation and apply an integrated approach for breakthrough innovations.”

Theo Tekstra from company sponsors Fluence emphasized the trade-off between the efficiency and spectra of LED lighting. “We can now mix different color combinations for different spectra. Don’t just look at the light but look at the combination of LEDs with other sensors,” Tekstra said. “When I look at how we innovate in the Netherlands compared to abroad, I see that we use a slightly different methodology but we are still leading in lighting innovation.”

Smart and sustainable farming needs a systems approach integrating all aspects of the crop production system. This starts with understanding the crop, obtaining interpretable information from sensors, and using models and AI to help growers around the globe produce healthy food for a growing population.