@inproceedings{Calandra:2015,
author = {Calandra, R and Ivaldi, S and Deisenroth, MP and Peters, J},
publisher = {IEEE},
title = {Learning Torque Control in Presence of Contacts using Tactile Sensing from Robot Skin},
url = {http://hdl.handle.net/10044/1/26653},
year = {2015}
}

Download

TY  - CPAPER
AB  - Whole-body control in unknown environments ischallenging: Unforeseen contacts with obstacles can lead topoor tracking performance and potential physical damages ofthe robot. Hence, a whole-body control approach for futurehumanoid robots in (partially) unknown environments needsto take contact sensing into account, e.g., by means of artificialskin. However, translating contacts from skin measurementsinto physically well-understood quantities can be problematicas the exact position and strength of the contact needs to beconverted into torques. In this paper, we suggest an alternativeapproach that directly learns the mapping from both skinand the joint state to torques. We propose to learn suchan inverse dynamics models with contacts using a mixtureof-contactsapproach that exploits the linear superimpositionof contact forces. The learned model can, making use ofuncalibrated tactile sensors, accurately predict the torquesneeded to compensate for the contact. As a result, tracking oftrajectories with obstacles and tactile contact can be executedmore accurately. We demonstrate on the humanoid robot iCubthat our approach improve the tracking error in presence ofdynamic contacts.
AU  - Calandra,R
AU  - Ivaldi,S
AU  - Deisenroth,MP
AU  - Peters,J
PB  - IEEE
PY  - 2015///
TI  - Learning Torque Control in Presence of Contacts using Tactile Sensing from Robot Skin
UR  - http://hdl.handle.net/10044/1/26653
ER  - 

Download