Adaptive Architectures for Transferability of Greenhouse Models

This project aims at a closer cooperation between the Biosystems Engineering Group of the Humboldt-Universit?t zu Berlin (HUB) and the BioRobotics Institute of the Scuola Superiore Sant’Anna (SSSA), Pisa, Italy. The international collaboration will explore the topic of adaptive architectures for the portability of greenhouse models.
The greenhouse production sector experiences a trend towards higher automation levels, aiming at performing precision horticulture by fine tuning the plants’ environment to better meet their needs and increase the production per unit area. Higher automation levels, however, face several challenges in biological environments, like: a) an inherent variability of biological organisms, b) an active role of plants, animals and humans in modifying their surroundings and c) high variability in time (by the weather, by plant growth). In the case of greenhouses, this traduces into differences between crops and varieties, greenhouse structures, climatic and plant (e.g. transpiration cooling) effects, to name a few.
The field of developmental robotics deals with similar problems in another context. Researchers draw inspiration from the mental processes occurring during development in humans, which allow them to deal with uncertainty and changing environments with highly plastic, while still robust, motor strategies, with the aim of producing adaptive behaviours in robots.
This collaboration will combine methods from developmental robotics and artificial intelligence (SSSA) for transferring models of plant photosynthesis (as main drive for gross yield production) between research facilities and production greenhouses (HUB). Aside from the basic scientific value of a novel transdisciplinary field of research, it also has a high potential to yield valuable control strategies for the plant production sector.
It is strived to investigate the suitability of adaptive and predictive deep neural models (SSSA) to relate (map) plant sensors and control actions in greenhouses (HUB). Learning these mappings would allow the system to anticipate (SSSA) the effects of intervention upon the crop and thus better plan further control actions (HUB).
The stay of Dr. Schillaci at HUB will aim at establishing the collaboration. Dr. Miranda will introduce greenhouse technologies and measurements, which will be analysed to set up the goals of an adaptive architecture for greenhouse models. The focus shall be a conceptual design to explore the potential of the adaptive models currently investigated by Dr. Schillaci's group. The subsequent stay of Dr. Miranda at SSSA will aim at discussing an experimental design to develop the adaptive and predictive models previously identified. Furthermore, both visiting periods will be exploited to lay the ground for a joint publication as well as for the definition of a strategy for further collaboration between SSSA and HUB, including the investigation of EU funding opportunities for the joint project.