AI-TEST-FIELD - A Concept for Industrial-Grade Development of Semantic Environment Perception

Jan Christoph Krause; Sebastian Röttgermann; Naeem Iqbal; Matthias Müter; Dominik Nieberg; Jaron Martinez; Mark Niemeyer; André Berghaus; Stefan Menke; Jens Herbers; Michael Kreyenhagen; Stefan Stiene; Arno Ruckelshausen; Joachim Hertzberg

In: Workshop on Agricultural Robotics and Automation. European Conference on Mobile Robots (ECMR-2021), ECMR 2021 Workshop on Agricultural Robotics and Automation, located at European Conference on Mobile Robots, August 31 - September 3, Bonn, Germany, unknown, 8/2021.


Agricultural environments are characterized by harsh and changing conditions. This results in high requirements regarding the robustness of sensors and algorithms for agricultural machines. Due to completely different environmental conditions, it is impossible to copy-paste sensors and algorithms from automotive engineering. From a research perspective, there are lots of algorithms that deal with an interpretation of a huge amount of sensor data, used for environment perception. Several well-known and high-performance algorithms are data driven. Therefore they need a lot of labeled training and evaluation data. To develop and evaluate a sensor system with focus on reliability, a bunch of different data sets – under different environmental conditions – have to be recorded. Possible conditions such as rain, dazzling sunlight, dust as well as different stages of plant growth have to be captured. Hence an environment is needed on which a lot of data can be generated in a simple way. The research project AI-TEST-FIELD aims an automated testbed taking into account the different environmental conditions over a long period of time.


Deutsches Forschungszentrum für Künstliche Intelligenz
German Research Center for Artificial Intelligence