On Friday 5 November, after a 24-hour hackathon in which 17 international teams took part, the five winning teams for the Autonomous Greenhouse Challenge were announced in a hybrid event. The third edition of the challenge is organized by Wageningen University & Research and Tencent. The five winning teams are CVA, digital_cucumber, MondayLettuce, VeggieMight, and team Koala. The winning teams will receive a place in the final challenge that will take place from February 2022 onwards. A lettuce crop then has to be grown with artificial intelligence algorithms - fully autonomous without human interaction - in reality.
The hackathon consisted of different parts: an international jury gave points to the teams for their AI approach (40% of the points) as well as the net profit that was achieved in a game to grow lettuce digitally, in a virtual greenhouse (60% of the points).
The jury selected 5 teams to proceed to the next round, in which they have to grow lettuce in a greenhouse compartment autonomously and in the real world. The 5 winning teams are:
- Team CVA won first place with 87 points. They ranked second place in the summer Online Challenge and once again showed a consistent performance in the hackathon. Team CVA – short for Crop Vision and Automation – comes from Korea and consists of team members of Gyenggi University of Science and Technology, Agri-Food Human Resource Development Institute Croft, Motion2AI, and Universal robots. Some members were also part of teams in the former edition.
- Team digital_cucumber were able to grow lettuce digitally and ranked second in the Hackathon. The team comes from Russia and has members of the Russian Agricultural Bank, HSE University, Lomonosov Moscow State University, Russian State Agrarian University, Moscow Timiryazev Agricultural Academy, and Moscow Institute of Physics and Technology.
- Team MondayLettuce was also from Korea, with members of Kangwon National University, Pusan National University, and ioCrops.
- Team VeggieMight has team members with different nationalities (coming from different Ukrainian companies, such as Quantum, HortiPolaris, Horticompass Agri-tech Consultancy, Robolect B.V., and students from Wageningen University and Research). They finished their task in the Hackathon the fastest.
- Team Koala won a wild card during the summer Online Challenge. This Online Challenge served as a preparation for the current Hackathon. It included optimized climate control and the development of computer vision algorithms for automated extraction of plant traits from lettuce images. The team captain of team 'Koala' is Kenneth Tran, who also led the winning team 'Sonoma' in the first edition of the Autonomous Greenhouse Challenge in 2019.
During the pitch, held for an international jury, teams presented their scientific and practical approach for fully autonomous control of a greenhouse in terms of climate and crop production. The teams received points (40%) for their AI approach (new, functional, robust, scalable, deployment of artificial intelligence).
More points (60%) could be gained in the game environment: the net profit realized in euros in virtual lettuce production. The teams were given access to a climate model and a lettuce crop growth model that was developed by researchers of WUR.
The organizers had ensured that the models had so many possibilities that a "gaming" situation arose. The crop sometimes reacted differently than expected in a practical situation. The teams themselves determined the operational decisions on operating screens: artificial lighting, heating, supplemental CO2, and crop spacing decisions. They did this using their own AI algorithms. They were growing crops virtually, but had to consider weather forecasts and could not redo the past, creating realistic circumstances.
From February 2022 onwards, the AI algorithms developed by these teams will be running lettuce cultivation in a real greenhouse compartment. Two cycles of a lettuce crop must then be grown fully autonomously. The first cycle is to test the algorithms and to obtain additional training data. After allowing the teams some time to refine their algorithms, if needed, the second cycle will start in May. The result of this second cycle determines the winner.
For more information:
Wageningen University & Research