We chose WUR because they are leading in cutting edge technologies for horticulture
23 January 2026
DENSO CORPORATION wants to make the impossible, possible by changing food systems from the inside out. In their ambition to do so with their tomato harvesting robot Artemy®, WUR plays a crucial and indispensable role.
Certhon, a DENSO group company based in the Netherlands, specialises in the design, development, and construction of custom-made climate systems, high-tech greenhouses, and indoor farms. Guided by its vision of ‘growing anything, anywhere, for everyone’, the company delivers solutions ranging from tomato production in the desert to growing lettuce in North America and strawberries in Hong Kong. The company strives to do so by changing food systems from the inside out, and provides local support for projects and developments.
Find the best experts in their domain
DENSO is an advanced global automotive parts manufacturer headquartered in Japan. Leveraging its core technologies, DENSO is expanding the scope of its value offerings into agriculture (agtech) and working to strengthen the agricultural industry. Since early September 2023, Certhon Group is wholly owned by DENSO. Backed by this, both Certhon and DENSO have the facilities, the knowhow, skills and financial ability to successfully develop, manufacture and market robots for horticulture. So why did the companies select Wageningen University & Research (WUR) to help them with their tomato harvesting robot Artemy?
Daigo Akutsu, responsible for technical business development, started the project at DENSO in Japan and consequently moved to Certhon in the Netherlands later. He clearly states the company’s motivation to cooperate with WUR. “DENSO is a technology group with a lot of companies in Japan and in other countries. But our main domain always has been automotive and it still is. When we are looking for partners for (collaborative) developments, we always carefully consider who are the best and preferred partners. We have set our vision as ‘Providing safe, delicious, and environmentally friendly food anytime, anywhere, for everyone.’ To realise this vision, we drive our business to enable stable and predictable production. We strive to cooperate with partners who share our vision and are the best experts in their respective domain(s). WUR is not only world-famous for their research and achievements in the agricultural and horticultural domains, they are also undoubtfully leading in cutting edge technologies for horticulture. We’ve now been collaborating on the Digital Twin Project since May 2024, and we’ve learned that they are very capable, have excellent knowledge of the domain, ánd are able to meet challenging deadlines.”
‘Nobody has ever done this before’
WUR’s expertise in crop modelling, robotic simulation and 3D modelling is used to accelerate the development of Certhon’s tomato harvesting robot Artemy. In this project, WUR aims to create a simulation where the Artemy robot is put in a virtual greenhouse. An important aspect is to also simulate the crop, including natural variation, realistic looks and accurate physical interaction between robot and crop. Aspects like crop modelling and growth simulation with virtual plants have been done by WUR plant scientists and specialists before, but the purpose was completely different, and life-like visual realism was never required.
Having a robot in simulation that can interact with realistic looking plants to simulate crop or fruit harvest, was a first, says Arjan Vroegop, researcher at WUR and part of the Vision+Robotics team. He is also leading the project on behalf of WUR. “While robotic simulation is quite common in controlled areas such as car factories, it is a novelty in the horticultural domain. Nobody has ever done this before. It is not a standard engineering project. Nor for DENSO or Certhon, nor for us. We are really working on the cutting edge of technology here.”
Akutsu adds: “When we started to collaborate and set mutual goals, nobody knew in advance if those goals were possible or realistic because nobody had ever done anything similar before. That makes it difficult to judge whether something is realistic to expect. We know how difficult research, development and innovation is. And considering the fact that this is our first project, our first challenge together, we could not expect too much. I therefore carefully documented my expectations of the project and the cooperation. Not only for myself and my team at the time we started this project in 2024, also for our top management and the Certhon team, and for Arjan and his team. Arjan immediately said ‘let’s try’ and I have to admit; I am impressed with their work and speed so far! The WUR experts are very pro-active and forward thinking, yet sometimes we expect too much. Therefore it is very important to constantly communicate and synchronise developments and progress, to manage mutual expectations.”
‘Top management was very impressed’
In March 2025, an intermediate gate review was done where Daigo Akutsu had to present the progress of the project to the top management. “I showed the robotic simulation achievements to our top management and they were impressed by the results achieved in such a short period of time (May 2024 – March 2025). Not only by the intermediate outcome, but also by the collaboration, the communication and the relationship built between the project partners. We were still facing challenges and had to meet some of the goals that were initially set. But the management realised that those goals were very ambitious without even knowing in advance what kind of work exactly needed to be done. By experience, we know that it sometimes takes a bit more time than expected or anticipated.”
Accelerate robotic developments
Akutsu explains: “Artemy is developed inhouse and during the development period, we did a lot of tests and evaluations in our greenhouse in Japan. This greenhouse is a bit of a drive from our office and you physically have to go there to run your tests. For those tests, we rely on vision technologies to detect and measure tomato plants in order to develop the harvesting technology suite for Artemy. We quickly learned that capturing plants is totally different from automating car factories… Conditions in greenhouses vary quickly and plants grow, develop and change. There’s seasonal aspects and influences, and after you’ve harvested a fruit, that situation will never be the same again. You cannot recreate the same conditions once they’ve changed. That made the evaluation and development of Artemy very time consuming and not as effective and fast as we wanted. We then asked ourselves the question, how to accelerate the development? The answer was by using simulation – as already implemented by our automotive division – in order to create reproduceable tests and reproduceable test environments and be able to retest the same situation after changing the robot and/or software.”
Artemy havesting robot (source: https://www.denso.com/global/en/driven-base/project/artemy/)
Vroegop adds: “Testing developments in real world situations takes a lot of time in the agricultural and horticultural domains because the conditions continuously change. If you are able to simulate not only the movements of the Artemy robot, but also robot-plant interaction, we can help speed up the development of the robot. This is where Wageningen comes in, as we have the different expertise required. Within Vision+Robotics, we have the expertise on robotics and simulation, but the team also consists of members of the WANDER XR Experience Lab and WUR’s Crop Physiology team for plant knowledge. The Wander XR Experience Lab has expertise on 3D modelling and game engines, and helped us with getting the looks of the tomato plants right. This makes WUR truly unique throughout the world. Having a plant model is one thing, we actually aim to make every plant realistic and unique. And this is where Maarten van der Meer (Crop Physiology) adds his expertise, because he knows all about plants.”
Creating virtual plants
“As a plant scientist in crop physiology I work on 3D crop modelling, where I model the functional and structural aspects of plants. In this particular project, these are cocktail tomatoes. We measure a number of plants and reconstruct them virtually based on logic and use Blender to create all the variation we want. By measuring real plants and determining key plant features and characteristics such as height, the lowest part of the leaves and other functional units of a plant referred to as phytomers, we make sure we create virtual plants that resemble natural variation. This involves randomisation within measured boundaries, while allowing for complex situations where crops are more ‘difficult’ to harvest the tomatoes of. We can then compare virtual plants based on a similar number of leaves, a similar number of fruits, et cetera. If you know to what extent plants vary and differ, you can create many different plants, which is a key requirement for the simulation.”
Besides a realistic visual environment, the simulation also provides semantic and instance segmentations (used for training detection models)
Van der Meer continues: “A critical and beneficiary aspect of the WUR approach is the measurement protocols we use. Because of this, we can feed our plant generator and our models to deal with different varieties as well. So we can shift from cherry tomatoes to cocktail tomatoes as it enables us to change the size, length and shape of the fruits we use in the models. They are all fully adaptable. This is crucial in relation to the DENSO developments as they focus on cherry tomatoes for now, but also may focus on other tomato varieties in future. We are now very close to having everything automated and I’m very much looking forward to seeing it in action to iteratively speed up the robot development.”
Not everything goes right instantly
So does the WUR-team succeed in doing things right every time? “I have a good example of something that initially went wrong”, says Vroegop. “Remember that this kind of modelling and simulation was never done before when we started. And that the developments in artificial intelligence (AI), machine learning and algorithms evolve at an incredibly rapid pace. So at the time we started to make virtual plants, we were limited in choosing from the available engines to do so. We decided to choose one of the main engines in use to make computer games. But we came to the conclusion that we underestimated the difficulty of plant physics, which we couldn’t get right. We then advised to switch to a dedicated robot simulation platform, with an automated workflow to import plants generated in Blender. This makes sure we can benefit from the latest and most advanced AI and machine learning technology and make it more future proof. I regard this as a typical example of the synergy we can achieve due to the vast amount of different expert researchers and scientists in Wageningen.”
Daigo Akutsu concludes: “In this project we face a lot of challenges but we face them together. At DENSO, our team had no prior experience with the engines Arjan mentioned, so after reviewing the details, we rely on and trust the advice and decisions by the WUR-team. The goal for this phase of the project is to be able to record synthetic data and automate the robot testing. That way, we can create a virtual Artemy tomato harvesting robot and use the outcomes to further improve the real Artemy robot. At DENSO, we are convinced that this cooperation will help us accelerate tomato growing and automated harvesting. So we can enable our existing and future customers to grow anything anywhere.”
