Automatische Wiedererkennung von individuellen Objekten mit dem Dynamic Anchoring Agent

Martin Günther, Friedemann Kammler, Oliver Ferdinand, Joachim Hertzberg, Oliver Thomas, Oliver Zielinski

In: HMD - Praxis der Wirtschaftsinformatik (HMD) XXX Springer 2020.


Individual objects play a central role in everyday life and make their distinction an important requirement for the application of robotics. The challenge is not only sensorial perception, but also the combination with existing knowledge: For example, two goods that are identical for object recognition can be further distinguished by their function ("The battery screwdriver I used last") or their environmental context ("The cup on my table"). Robotics addresses this challenge under the notion of anchoring, meaning the ability to identify individual objects and to recognize them by their context once they have left the perception of the system. On a technical level, the problem is to establish relationships between sensor data and symbols in a knowledge base and thus to address physical objects concretely. This paper presents the anchoring problem and the Dynamic Anchoring Agent (DAA) as a solution approach. The use of the DAA is demonstrated using two real application scenarios: With the example of a MakerSpace, the possibilities for extended cooperation between humans and robots are shown - for instance in the search and identification of personal tools or required production materials. The second application demonstrates the extended use in dynamic environments within the example of a marina.


German Research Center for Artificial Intelligence
Deutsches Forschungszentrum für Künstliche Intelligenz