S3K: Self-Supervised Semantic Keypoints for Robotic Manipulation via Multi-View Consistency

Mel Vecerik, Jean-Baptiste Regli, Oleg Sushkov, David Barker, Rugile Pevceviciute, Thomas Rothörl, Raia Hadsell, Lourdes Agapito, Jonathan Scholz
Proceedings of the 2020 Conference on Robot Learning, PMLR 155:449-460, 2021.

Abstract

A robot’s ability to act is fundamentally constrained by what it can perceive. Many existing approaches to visual representation learning utilize general-purpose training criteria, e.g. image reconstruction, smoothness in latent space, or usefulness for control, or else make use of large datasets annotated with specific features (bounding boxes, segmentations, etc.). However, both approaches often struggle to capture the fine-detail required for precision tasks on specific objects, e.g. grasping and mating a plug and socket. We argue that these difficulties arise from a lack of geometric structure in these models. In this work we advocate semantic 3D keypoints as a visual representation, and present a semi-supervised training objective that can allow instance or category-level keypoints to be trained to 1-5 millimeter-accuracy with minimal supervision. Furthermore, unlike local texture-based approaches, our model integrates contextual information from a large area and is therefore robust to occlusion, noise, and lack of discernible texture. We demonstrate that this ability to locate semantic keypoints enables high level scripting of human understandable behaviours. Finally we show that these keypoints provide a good way to define reward functions for reinforcement learning and area good representation for training agents.

Cite this Paper

BibTeX
@InProceedings{pmlr-v155-vecerik21a, title = {S3K: Self-Supervised Semantic Keypoints for Robotic Manipulation via Multi-View Consistency}, author = {Vecerik, Mel and Regli, Jean-Baptiste and Sushkov, Oleg and Barker, David and Pevceviciute, Rugile and Roth\"orl, Thomas and Hadsell, Raia and Agapito, Lourdes and Scholz, Jonathan}, booktitle = {Proceedings of the 2020 Conference on Robot Learning}, pages = {449--460}, year = {2021}, editor = {Kober, Jens and Ramos, Fabio and Tomlin, Claire}, volume = {155}, series = {Proceedings of Machine Learning Research}, month = {16--18 Nov}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v155/vecerik21a/vecerik21a.pdf}, url = {https://proceedings.mlr.press/v155/vecerik21a.html}, abstract = {A robot’s ability to act is fundamentally constrained by what it can perceive. Many existing approaches to visual representation learning utilize general-purpose training criteria, e.g. image reconstruction, smoothness in latent space, or usefulness for control, or else make use of large datasets annotated with specific features (bounding boxes, segmentations, etc.). However, both approaches often struggle to capture the fine-detail required for precision tasks on specific objects, e.g. grasping and mating a plug and socket. We argue that these difficulties arise from a lack of geometric structure in these models. In this work we advocate semantic 3D keypoints as a visual representation, and present a semi-supervised training objective that can allow instance or category-level keypoints to be trained to 1-5 millimeter-accuracy with minimal supervision. Furthermore, unlike local texture-based approaches, our model integrates contextual information from a large area and is therefore robust to occlusion, noise, and lack of discernible texture. We demonstrate that this ability to locate semantic keypoints enables high level scripting of human understandable behaviours. Finally we show that these keypoints provide a good way to define reward functions for reinforcement learning and area good representation for training agents.} }
Endnote
%0 Conference Paper %T S3K: Self-Supervised Semantic Keypoints for Robotic Manipulation via Multi-View Consistency %A Mel Vecerik %A Jean-Baptiste Regli %A Oleg Sushkov %A David Barker %A Rugile Pevceviciute %A Thomas Rothörl %A Raia Hadsell %A Lourdes Agapito %A Jonathan Scholz %B Proceedings of the 2020 Conference on Robot Learning %C Proceedings of Machine Learning Research %D 2021 %E Jens Kober %E Fabio Ramos %E Claire Tomlin %F pmlr-v155-vecerik21a %I PMLR %P 449--460 %U https://proceedings.mlr.press/v155/vecerik21a.html %V 155 %X A robot’s ability to act is fundamentally constrained by what it can perceive. Many existing approaches to visual representation learning utilize general-purpose training criteria, e.g. image reconstruction, smoothness in latent space, or usefulness for control, or else make use of large datasets annotated with specific features (bounding boxes, segmentations, etc.). However, both approaches often struggle to capture the fine-detail required for precision tasks on specific objects, e.g. grasping and mating a plug and socket. We argue that these difficulties arise from a lack of geometric structure in these models. In this work we advocate semantic 3D keypoints as a visual representation, and present a semi-supervised training objective that can allow instance or category-level keypoints to be trained to 1-5 millimeter-accuracy with minimal supervision. Furthermore, unlike local texture-based approaches, our model integrates contextual information from a large area and is therefore robust to occlusion, noise, and lack of discernible texture. We demonstrate that this ability to locate semantic keypoints enables high level scripting of human understandable behaviours. Finally we show that these keypoints provide a good way to define reward functions for reinforcement learning and area good representation for training agents.
APA
Vecerik, M., Regli, J., Sushkov, O., Barker, D., Pevceviciute, R., Rothörl, T., Hadsell, R., Agapito, L. & Scholz, J.. (2021). S3K: Self-Supervised Semantic Keypoints for Robotic Manipulation via Multi-View Consistency. Proceedings of the 2020 Conference on Robot Learning, in Proceedings of Machine Learning Research 155:449-460 Available from https://proceedings.mlr.press/v155/vecerik21a.html.

Related Material