Haptic Inspection of Planetary Soils With Legged Robots

Planetary exploration robots encounter challenging terrain during operation. Vision-based approaches have failed to reliably predict soil characteristics in the past, making it necessary to probe the terrain tactilely. We present a robust, haptic inspection approach for a variety of fine, granular m...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:IEEE robotics and automation letters Jg. 4; H. 2; S. 1626 - 1632
Hauptverfasser: Kolvenbach, Hendrik, Bartschi, Christian, Wellhausen, Lorenz, Grandia, Ruben, Hutter, Marco
Format: Journal Article
Sprache:Englisch
Veröffentlicht: Piscataway IEEE 01.04.2019
The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
Schlagworte:
AI-based methods
Classification
Discrete Wavelet Transform
Foot
Force and tactile sensing
Granular materials
Granular media
Inspection
Legged locomotion
legged robots
Robot sensing systems
Robots
Sensors
Soil
Soils
Space exploration
space robotics and automation
Support vector machines
Terrain
Wavelet transforms
ISSN:2377-3766
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:Planetary exploration robots encounter challenging terrain during operation. Vision-based approaches have failed to reliably predict soil characteristics in the past, making it necessary to probe the terrain tactilely. We present a robust, haptic inspection approach for a variety of fine, granular media, which are representative of Martian soil. In our approach, the robot uses one limb to perform an impact trajectory, while supporting the main body with the remaining three legs. The resulting vibration, which is recorded by sensors placed in the foot, is decomposed using the discrete wavelet transform and assigned a soil class by a support vector machine. We tested two foot designs and validated the robustness of this approach through the extensive use of an open-source dataset, which we recorded on a specially designed single-foot testbed. A remarkable overall classification accuracy of more than 98% could be achieved despite various introduced disturbances. The contributions of different sensors to the classification performance are evaluated. Finally, we test the generalization performance on unknown soils and show that the interaction behavior can be anticipated.
Bibliographie:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 14
ISSN:2377-3766
DOI:10.1109/LRA.2019.2896732