Concept of a Test Environment for the Automated Evaluation of Algorithms for Robust and Reliable Environment Perception

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

In: Barbara Sturm, Henning Meyer (Hrsg.). LAND.TECHNIK 2022 The Forum for Agricultural Engineering Innovations. International Conference Agricultural Engineering (LAND.TECHNIK-2022) February 25-25 virtuell Germany ISBN 978-3-18-092395-6 VDI Verlag GmbH Düsseldorf 2/2022.


In agriculture processes harsh and uncertain surroundings conflict with a reliable interpretation of the environment by conventional sensor systems. Influences like rain, dust, dazzling light or variety of plant and soil characteristics lead to the need of robust methods for environment perception. To make the journey from research prototypes to sensor systems suitable for in-dustry, an evaluation environment is built up within the BMEL-funded research project AI-TEST-FIELD. In this paper the concept of a rail-based carrier system is described. For designing an auto-mated testbed, a model of the carrier is created within a simulation environment. Different scenarios and objects that are typical for agricultural settings were evaluated in a survey. To develop and evaluate AI-based algorithms for multiple sensors with different measurement principles, a machine-learning pipeline was added. To automatically find ground truth and label the sensor data with as little manual work as possible, surveying of the test field is needed.


Deutsches Forschungszentrum für Künstliche Intelligenz
German Research Center for Artificial Intelligence