Publikation
Field Trials and Evaluation Methods for Sensors used in Perception Systems for Autonomous Machinery with Respect to SOTIF
Jan Christoph Krause; Sebastian Röttgermann; Alexander Tauber; Jens Herbers; Stefan Menke; Mark Niemeyer; Lutz Plagge; Benedikt Thy; Hannes Hollmeier; Stefan Stiene; Joachim Hertzberg
In: LAND.TECHNIK AgEng 2024. International Conference on Agricultural Engineering (AgEng-2024), November 6-7, Osnabrück, VDI Verlag, 2024.
Zusammenfassung
Since experiments in agriculture are seasonally limited, development and evaluation of perception
systems require special planning and preparations. In the research project AI-TEST-FIELD a test
environment for the evaluation of sensor systems that offer the opportunity to generate precise
and safe operations in a wide range of environmental conditions, was developed. The test
environment combines a field-based test track, with which comparable sensor data is recorded
automatically, and an evaluation pipeline, in which different sub-datasets can be evaluated by
multiple algorithms for certain environmental conditions.
In connection with different test parameters an evaluation procedure is proposed. Based on an
analysis of triggering conditions for system insufficiency, test parameters and scenarios were
defined. These test cases are processed in individual as well as continuous long-term
experiments. Since tests on an abstract test environment are only one pillar for evaluation of a
complete perception system a transfer onto real agricultural machines is done during fieldwork
examples (e.g. mowing, tillage). The possibility to transfer outcomes from the test field
environment to the real autonomous machines is one important aspect for the industry partners
Maschinenfabrik Bernard Krone GmbH & Co. KG and LEMKEN GmbH & Co. KG within their
development project Combined Powers.