Sissy Friedrich, Adrian Bauer, Daniel Leidner of the FUTURO YIG and David Spieler from Munich University of Applied Sciences present their work on context-aware contact classification using physics-based semantic knowledge on the IROS 2021 Workshop on Impact Aware Robotics.

Sissy “working” with the robot Rollin’ Justin. Credit: DLR/Sissy Friedrich

It is inevitable that at some point a robot will encounter undesired collisions with a human. On the other hand, a robot has to intentionally make contact in order to interact with the environment. Accordingly, undesired collision events have to be distinguished from intentional contact during purposeful interaction with the environment. To make this distinction, we propose an on-line classification approach that uses labels generated from off-line physics simulations, which offloads computationally intensive processes while semantic information is preserved. Following this automated labeling process, a classifier is trained on time series data recorded from various contact and collision scenarios with the humanoid robot Rollin’ Justin. The evaluation shows that our approach is able to distinguish between different contact and collision situations at runtime, allowing for safe collaboration without constraining the execution of contact-rich tasks.

On-line classification of different contact types. Credit: DLR

This concludes the stay of Sissy at DLR after 1.5 years in which she successfully completed her Master of Applied Research in Engineering Sciences (MAPR) with distinction. A full article on the work will follow in the comming months.