CP 4: Autonomous In-Field Intervention

With this project we pursue two core developments: Autonomous robotic weeding at the individual plant level and precise application of nitrogen (N) fertilizer enabled by digital avatars performing agro-ecosystem simulation and predictions.

Precise robotic weeding aims to intervene in a minimally invasive manner reducing the amount of inputs like herbicides but also to selectively choose which weeds have to be managed at a particular time and how they are managed. Therefore, we develop an autonomous field robot prototype, which can detect and identify individual plants in real-time in the field. Identifying the weed species, or biological sub-group, with modern learning algorithms enables the robot prototype to take the most appropriate intervention method at each individual plant, which can be chemicals, thermal, mechanical intervention, or no intervention. Through these alternative minimally invasive intervention techniques, herbicide resistance in key species can be prevented. Furthermore, we create several weed maps per season to observe species, plant size and growth rate and analyze the trade-off between crop productivity (yield) and biodiversity.

The second application of the robotic technology applies to fertilization in combination with Digital Agricultural Avatars (DAA) that inform robots and the next generation of tractors to address crop nitrogen deficiency based on information received from in situ observations from ground or airborne autonomous systems. Such digital avatars are based on a hierarchical model system based on the TerrSysMP platform and process-based crop growth models.

We envision new unmanned ground vehicles to determine the N status and to derive meter-scale state variables to inform and validate the modeling system. In this way, deeper insights in potential remotely sensed observables with respect to soil-plant interactions are analyzed to help to describe stress impacts but also to determine the fertilization status to optimized nitrogen fertilization strategies. Overall, these findings finally support the development of meter-scale DAAs useable to improve forecasts of water and fertilizer induced stress. The forecast time scales will range from days to seasons and include all relevant variables of agricultural systems. The forecast products are utilized online to provide intervention strategies for optimal application of fertilizer doses depending on the soil and environmental status of the sub-field plots. These can be used to precisely apply nitrogen fertilizer closing. This completes the loop from observations to intervention, all in real time in the field.