agrilution: Letting kitchen herbs grow with hard- and software
Munich-based company agrilution started in 2013 as one of the pioneers in vertical farming consumer products. Its centerpiece is called plantCube, an intelligent plant box for the cultivation of herbs and salads at home, for which they recently won a Green Product Award. I discussed their technological approach with Fabian Göckel (28), who is leading agrilution’s software team.
Q: Fabian, tell me a little bit about yourself. How did you come to join agrilution?
Fabian: After my graduation at TU Munich, I was working as a developer in the automotive industry. But I was looking for a job with faster innovation cycles. Since I am very passionate about home growing, I already did some just-for-fun experiments with automated irrigation of my own plants. So when I saw the startup agrilution is hiring software engineers, I immediately applied for the job. That was back in 2016 and I think I was about staff member number eight. Today, we have over 25 employees in total with four full-time developers in our software department.
Q: Sounds like at agrilution you are taking software very seriously?
Fabian: Absolutely. We have developed a unique combination of hardware, control software, and an intelligent app. That’s what’s so exciting about it. Personally, I’m fascinated by how the plantCube has challenged me to build and program the device in order to maximize plant growth through automatic irrigation, perfect lighting (we use Osram Plant Light LED), and precise climate control. The customer is able to observe the plants’ daily growth from sowing the seeds to harvest, all while the plantCube takes care of everything. It combines various aspects from all software engineering areas, such as the app’s user-friendly interface or the suitable backend for live optimization and long-term analysis of the collected data.
“We had to build everything from scratch”
Q: We all know this does not happen overnight. How did you get started?
Fabian: Our founder Max Lössl first thought of a smart home appliance sized vertical farm on a trip around the world seven years ago. Together with Philip Wagner, he founded agrilution in early 2013 and our first software prototype for the system was developed with an external service provider. However, it became obvious that software will play a crucial role in our future product roadmap. And as there are so many things changing while developing a complex product, we needed a more flexible and agile approach. On basis of the existing prototype, we have built an in-house software team and formed our own holistic system for the plantCube. The product is a stable ecosystem with all prerequisites to establish itself as a smart home appliance in the near future. We are also using existing platforms for the IoT stuff, however there were not any farming specific frameworks that we could build on. So we had to develop everything on our own from scratch.
Q: What have been the biggest challenges in this respect?
Fabian: It is always hard to predict the future and set up a system that is prepared for all potential requirements of the next 2–5 years. Regarding software development, we try to tackle this by being as modular as possible, so we can upgrade specific elements. There is also a constant challenge of new components on the market and a natural conflict between — on the one hand — being innovative and — on the other hand — providing a 100% stable product for the customer to grow his plants.
Q: …and I guess it requires a good collaboration between your departments?
Fabian: Yes, that is one key to success. In our lab, we have developed innovative algorithms, which constantly optimise the growth of each individual plant variety in the plantCube. In order to achieve this, interdisciplinary cooperation within the team is necessary, as biological know-how regarding light-, temperature-, and irrigation-settings are intelligently incorporated into the code. The collected insights are combined into so-called plant recipes. These are defined in terms of plant properties such as taste, nutrient content, yield, appearance and texture. The main focus currently lies on each individual plant’s taste in its different growth stages.
“We are not interested in who uses the plantCube. Only how it is being used.”
Q: Isn’t there sometimes conflict potential between hardware, software and plant experts?
Fabian: Definitely! When it comes to the parametrization of our system, a lot of different aspects come together and finding the optimal solution can be a tedious job. I am giving you an example: For our temperature control we adapt the amount of air, which is led through the cooling fins based on the difference between measured and target temperature. As the plant scientist wants to have a stable environment for all plants, he is interested in a quickly reacting control function. This would lead to a more or less continuously active cooling, which again is in contrast to the goals of our product engineers, who try to minimize power consumption and decrease the workload of all components. If you would go for a control with smaller cycles, but higher activity, the overall noise level gets to high for an end consumer product. So we as software engineers try to bring all those aspects together and aim for the overall optimal solution.
Q: That makes sense. What kind of sensor data do you gather in the plantCube?
Fabian: To control the environmental parameters, the plantCube is equipped with air sensors for temperature and moisture, as well as water sensors for EC and temperature. We are also working with optical and depth sensors, which allow us to measure for example growth rate and to track the form of growth for each plant. Data like these is not yet used in a direct control loop, but is evaluated by our plant experts in order to monitor and optimize the recipe algorithms. When the time comes, our software systems are ready to run multiple variations of our algorithms, adapting themselves according to the measured performance data of past growth cycles.
Q: And how is the sensor data processed?
Fabian: We have integrated two microcontroller chips, one for the sensor and actuator controls, one for connecting the system to the cloud. The plantCube automatically detects via RFID which type of plants are placed in the device. Then our plant recipes are transferred from the cloud to the actual device to control the optimal conditions during the different growing stages. Sometimes, the best actuator is the user. So he gets notifications to change the water, the nutrient cartridge or his seed mats. All temperature related things are controlled through automated regulation cycles. In order to guarantee seamless interaction with the plantCube, we use a fail-safe cloud platform with secure and redundant servers, located in Germany. We attempt to keep all algorithms on the cloud in general. This ensures a high level of adaptability. In the worst case, the plantCube is designed to flawlessly process all recipes for at least one month without internet connection. Data sensitivity continuously stands at the forefront of all developments, this is why all connections and data are encrypted or anonymized, meaning no conclusions can be drawn about a specific end user. We are not interested in who uses the plantCube, only how it is being used.
“At the end of the day, taste is the most important thing”
Q: And how will this feedback data be used to improve your algorithms?
Fabian: The plantCube offers a new set of opportunities for the future: we are now able to understand both the plant’s growth and the user’s behaviour much better. The resulting empirical values are implemented in new development phases, which creates an incredibly exciting and unexpected interface between biological and technological expertise.
Q: Your users are not professional farmers, how does this influence your product?
Fabian: Simplicity makes the difference. The app is a very important interface between the plantCube and our users. Our intuitive step-by-step instructions help even those with no green thumb to speak of, and those who consider themselves non technically-inclined, to easily use our product. I sit down with the UX & UI designer on a daily basis, together we work on the interface’s graphic elements and constantly optimize the usability. We can also use this channel to gather information about the taste and state of the plants from the users through a gamification approach. At the end of the day, taste is the most important thing.
Q: What future developments do you see in vertical farming technology?
Fabian: agrilution clearly sees a trend developing in the interconnection of systems and processes, a central data management and the resulting continuous parameter optimization. I think for a standardized horticulture software solution, existing systems have to be easily integratable, including custom hardware. Furthermore, a flexible parameterization of the functions, access to the raw data as well as simple options for the evaluation of all data is decisive to benefit the standardization.
We believe that ten years from now, the plantCube will be a self-learning device performing various micro-experiments as part of a swarm network in order to optimize itself. You’ll hear the concept “personalised food” come up often. Up until now, this has been a future vision of a diet that is individually tailored to the end user. I’m so strongly invested in agrilution, because I want every home to have access to a healthy diet — as fresh and crisp as never before!
I want to thank Fabian and agrilution for taking their time and providing these insights. There is no financial obligation of any form between agrilution and me, I publish this interview because I am a urban farming and software enthusiast. If you are interested in plantCube, you can pre-order it on the agrilution website (available in Germany only).
