Autonomous Greenhouses

Plant physiology aims to contribute to the understanding of how plants function, such as the processes that happen in plants in response to changes in their environment and how they adapt to stresses.

Environmental factors, such as light, have an effect on crop growth and yield. We use simulation models to research these effects in crops grown in greenhouses and vertical farms.

The application of physical models is a way to predict the greenhouse climate based on the principles of calculating heat and mass balances. Its components, such as absorption of solar and artificial light, heating, crop transpiration, condensation, and ventilation, follow well-known laws of physics. With computations based on these highly predictable processes, we can define and implement resource-use efficient climate control strategies.

Digital twins use data-integration, artificial intelligence (AI), and machine learning to create a virtual version of a crop. These simulation models are continuously fed real-time information from the actual circumstances in a greenhouse, making it possible to analyse and simulate processes and reactions of plants more accurately. With the help of digital twins, we can better monitor plants and predict future scenarios.

Multi- and hyperspectral imaging methods are some of the most powerful to measure constituents, stress, and disease symptoms in plants. Spectral imaging offers a wealth of information to detect objects and their location thanks to not only one, but a whole spectrum of values available per pixel. Using spectral imaging in combination with models, we can measure the spatial distribution of dry matter content, nitrogen status or sugar concentration, and even lycopene or chlorophyll concentration.

Non-invasive sensors can measure and monitor a variety of plant traits and climate conditions in greenhouses. The data provides insights to constantly improve the conditions in the greenhouse and support management decisions. Sensors have also shown potential for detecting plant diseases.

Robotic systems in greenhouses need to be able to work in a highly challenging environment and deal with complex products that show a lot of variation and change in time (grow or ripen, for example). We design robotic systems combining hardware and software to a functioning robot that can think, sense, and act in such conditions.

The learning capacity of a robot is a key part of their intelligence. It is possible to discover patterns using machine learning, for instance in photographs. When given a large number of training data sets, the artificial intelligence automatically learns how to teach itself not only what is in the image but also where it is. The ability of machine learning to handle the high degree of diversity found in natural products is a key strength. Machine learning can also be used, for example, to teach a robot to grip objects or determine yield predictions.

How do robots or other automated systems see data and turn it into actionable information? We can accurately analyse data to shape models, and models to objective information, using computer vision techniques. By assessing characteristics such as product size, surface flaws, leaf area, or stem length, we are able to automatically recognise different plant attributes (phenotyping).