Implementation of an On-Board Terrain Classfier Based on Propioceptive Sensor Data for a Planetary Rover
Raúl Domínguez, Lennart Kuhr, Jonathan Babel, Florian Cordes, Giulio Reina, Frank Kirchner
Editors: ,
In 16th Symposium on Advanced Space Technologies in Robotics and Automation, (ASTRA-2022), 01.6.-02.6.2022, ESA-ESTEC, Noordwijk, the Netherlands, ESA Conference Bureau / ATPI Corporate Events, Jun/2022.

Abstract :

The implementation of a Support Vector Machine based terrain classifier for the hybrid locomotion rover Sher- paTT is presented. In the first phase of classification the physical characteristics of the traversed terrain are statis- tically derived from proprioceptive data (i.e. engineered features). The features are then used by the classifier to distinguish between three different surface types: sand, compact sand and concrete. Based on previous offline studies [7] the terrain classifier has been integrated into the control architecture of the rover, as well as deployed and tested in analog environment. The software compo- nent runs completely on the on-board computer (OBC) of SherpaTT, embedded within the Robotics Construc- tion Toolkit (Rock)1 framework. Insights on the imple- mentation and the software architecture surrounding the classifier are provided. Performance metrics demonstrate that the terrain classifier can run in parallel with the rest of the control software on the OBC, achieving an overall high accuracy in terrain classification.



Files:

sherpatt_terrain_classifier_astra2022.pdf


© DFKI GmbH
last updated 06.09.2016
to top