Informationsverarbeitung mit Maschinellem Lernen für Taktile Sensoren

Publikationstyp
Conference Paper
Date Issued
2019
Sprache
German
Author(s)
Kühn, Jan  
Jäger, Joas  
Kuhl, Matthias  orcid-logo
Manoli, Yiannos  
Institut
Integrierte Schaltungen E-9  
TORE-URI
http://hdl.handle.net/11420/8129
Start Page
518
End Page
521
Citation
MikroSystemTechnik Kongress, Mikroelektronik (MEMS 2019)
Contribution to Conference
MikroSystemTechnik Kongress, Mikroelektronik MEMS 2019  
Scopus ID
2-s2.0-85096741609
Tactile fingers provide robotic gripping systems with essential sensor feedback for the manipulation of objects. The presented tactile sensor finger measures static forces and slip vibrations with an array of identical stress sensors. Instead of calibrating the sensor to measure forces quantitatively, machine learning is used to extract abstract information from the raw data of the sensors. This information processing scheme can classify the direction of applied forces with an accuracy of 99.8 % through the spatial distribution of stress sensors. At the same time, with the same signal processing, slip can be detected without spectral analysis with an accuracy of 99.6 %.