MidasTouch: Monte-Carlo inference over distributions across sliding touch

Sudharshan Suresh, Zilin Si, Stuart Anderson, Michael Kaess, Mustafa Mukadam
Proceedings of The 6th Conference on Robot Learning, PMLR 205:319-331, 2023.

Abstract

We present MidasTouch, a tactile perception system for online global localization of a vision-based touch sensor sliding on an object surface. This framework takes in posed tactile images over time, and outputs an evolving distribution of sensor pose on the object’s surface, without the need for visual priors. Our key insight is to estimate local surface geometry with tactile sensing, learn a compact representation for it, and disambiguate these signals over a long time horizon. The backbone of MidasTouch is a Monte-Carlo particle filter, with a measurement model based on a tactile code network learned from tactile simulation. This network, inspired by LIDAR place recognition, compactly summarizes local surface geometries. These generated codes are efficiently compared against a precomputed tactile codebook per-object, to update the pose distribution. We further release the YCB-Slide dataset of real-world and simulated forceful sliding interactions between a vision-based tactile sensor and standard YCB objects. While single-touch localization can be inherently ambiguous, we can quickly localize our sensor by traversing salient surface geometries. Project page: https://suddhu.github.io/midastouch-tactile/

Cite this Paper

BibTeX
@InProceedings{pmlr-v205-suresh23a, title = {MidasTouch: Monte-Carlo inference over distributions across sliding touch}, author = {Suresh, Sudharshan and Si, Zilin and Anderson, Stuart and Kaess, Michael and Mukadam, Mustafa}, booktitle = {Proceedings of The 6th Conference on Robot Learning}, pages = {319--331}, year = {2023}, editor = {Liu, Karen and Kulic, Dana and Ichnowski, Jeff}, volume = {205}, series = {Proceedings of Machine Learning Research}, month = {14--18 Dec}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v205/suresh23a/suresh23a.pdf}, url = {https://proceedings.mlr.press/v205/suresh23a.html}, abstract = {We present MidasTouch, a tactile perception system for online global localization of a vision-based touch sensor sliding on an object surface. This framework takes in posed tactile images over time, and outputs an evolving distribution of sensor pose on the object’s surface, without the need for visual priors. Our key insight is to estimate local surface geometry with tactile sensing, learn a compact representation for it, and disambiguate these signals over a long time horizon. The backbone of MidasTouch is a Monte-Carlo particle filter, with a measurement model based on a tactile code network learned from tactile simulation. This network, inspired by LIDAR place recognition, compactly summarizes local surface geometries. These generated codes are efficiently compared against a precomputed tactile codebook per-object, to update the pose distribution. We further release the YCB-Slide dataset of real-world and simulated forceful sliding interactions between a vision-based tactile sensor and standard YCB objects. While single-touch localization can be inherently ambiguous, we can quickly localize our sensor by traversing salient surface geometries. Project page: https://suddhu.github.io/midastouch-tactile/} }
Endnote
%0 Conference Paper %T MidasTouch: Monte-Carlo inference over distributions across sliding touch %A Sudharshan Suresh %A Zilin Si %A Stuart Anderson %A Michael Kaess %A Mustafa Mukadam %B Proceedings of The 6th Conference on Robot Learning %C Proceedings of Machine Learning Research %D 2023 %E Karen Liu %E Dana Kulic %E Jeff Ichnowski %F pmlr-v205-suresh23a %I PMLR %P 319--331 %U https://proceedings.mlr.press/v205/suresh23a.html %V 205 %X We present MidasTouch, a tactile perception system for online global localization of a vision-based touch sensor sliding on an object surface. This framework takes in posed tactile images over time, and outputs an evolving distribution of sensor pose on the object’s surface, without the need for visual priors. Our key insight is to estimate local surface geometry with tactile sensing, learn a compact representation for it, and disambiguate these signals over a long time horizon. The backbone of MidasTouch is a Monte-Carlo particle filter, with a measurement model based on a tactile code network learned from tactile simulation. This network, inspired by LIDAR place recognition, compactly summarizes local surface geometries. These generated codes are efficiently compared against a precomputed tactile codebook per-object, to update the pose distribution. We further release the YCB-Slide dataset of real-world and simulated forceful sliding interactions between a vision-based tactile sensor and standard YCB objects. While single-touch localization can be inherently ambiguous, we can quickly localize our sensor by traversing salient surface geometries. Project page: https://suddhu.github.io/midastouch-tactile/
APA
Suresh, S., Si, Z., Anderson, S., Kaess, M. & Mukadam, M.. (2023). MidasTouch: Monte-Carlo inference over distributions across sliding touch. Proceedings of The 6th Conference on Robot Learning, in Proceedings of Machine Learning Research 205:319-331 Available from https://proceedings.mlr.press/v205/suresh23a.html.

Related Material