AGROS II: Next steps towards an autonomous greenhouse

Greenhouse horticulture is facing a number of challenges. Companies are expanding, and contracts with suppliers are growing more complicated. At the same time, fluctuating energy prices and finding qualified workers remain a constant concern. For growers, it is becoming nearly impossible to oversee all these factors, even though they are critical for running a profitable business. This creates a clear need for systems that support growers in their decision making. In AGROS II - the 2nd stage of the AGROS project - Wageningen University & Research is working with industry and network partners to develop intelligent algorithms that automatically monitor the crop, irrigation, and climate, detect deviations and ensure cultivation runs according to plan.

The ambition: a fully autonomous greenhouse

"Our long-term goal is a fully autonomous greenhouse, where irrigation and climate are controlled based on sensor data and models, without grower intervention. The models will also indicate when crop management such as leaf pruning and fruit thinning should take place," says project leader Anja Dieleman. “An essential component of autonomous greenhouse control is the availability of continuous data on the crop, substrate and climate. In AGROSII, we are developing the sensors and methods to make this possible.”

From manual measurement to smart image analysis

Climate data such as temperature, light intensity and CO2 concentration have been measured with sensors for many years. Crop monitoring, however, is still done manually. This is time-consuming, prone to interpretation differences and limited to a small number of plants per greenhouse. To improve this, AGROS II works with an innovative camera platform that performs crop measurements automatically, more frequently and in a standardized manner. The platform collects images in a cucumber crop, from which key crop traits can be derived. These data a provide the models with the information they need to adjust the cultivation process when necessary.

Two routes to autonomous control: Digital Twin and AI

“Two approaches have been chosen to achieve autonomous greenhouse control: a Digital Twin and an AI algorithm. The Digital Twin is generated from WUR crop and climate models,” Dieleman explains. “This Digital Twin continuously simulates how the greenhouse climate and a virtual cucumber crop respond to different control actions. Based on these simulations, the Twin determines the optimal strategy. It receives near-real-time sensor data from the greenhouse and uses this to update itself, creating a system that learns from actual conditions and becomes more accurate over time.” 
In the second approach, an AI algorithm is trained by running thousands of simulations with a virtual cucumber crop, evaluating the effects of different control actions. Through this process, the model learns which actions lead to desirable outcomes and which do not. Once trained, the model can act autonomously based on current data from the greenhouse.

Data-driven cultivation 

In addition, a third cucumber crop is being managed in a data-driven way by a group of experienced cultivation experts. Using a tailor made dashboard, they can access all climate, substrate and crop data and provide cultivation advice accordingly. By comparing the digital overview with what they observe in the greenhouse, they can refine both their insights and the dashboard itself.

Soft sensors: greater data reliability

Reliable data are essential for autonomous cultivation control. AGROS II therefore also develops "soft sensors": smart combinations of sensor data and model calculations. This additional layer of verification significantly increases the reliability of the control system.

A widely supported sector initiative

AGROS II helps the greenhouse horticulture sector advance towards sustainable, fossil-free and circular cultivation systems. 
The project is carried out with a broad group of partners. Companies such as Nunhems/BASF Vegetable Seeds, Mechatronix, Gavita, and Cultilene contribute specialized knowledge of cultivation systems. The camera platform has been developed in collaboration with VDL ETG, while Delphy, Van der Hoeven and Source.ag provide expertise in data-driven cultivation. Stichting KIJK, Greenport West Holland, the Municipality of Lansingerland, and the Hagelunie Innovation Fund strengthen the project through grower engagement, societal input and a large network of horticultural entrepreneurs.