Graph-Structured Visual Imitation

Maximilian Sieb, Zhou Xian, Audrey Huang, Oliver Kroemer, Katerina Fragkiadaki
Proceedings of the Conference on Robot Learning, PMLR 100:979-989, 2020.

Abstract

We cast visual imitation as a visual correspondence problem. Our robotic agent is rewarded when its actions result in better matching of relative spatial configurations for corresponding visual entities detected in its workspace and the teacher’s demonstration. We build upon recent advances in Computer Vision, such as human finger keypoint detectors, object detectors trained on-the-fly with synthetic augmentations, and point detectors supervised by viewpoint changes [1] and learn multiple visual entity detectors for each demonstration without human annotations or robot interactions. We empirically show that the proposed factorized visual representations of entities and their spatial arrangements drive successful imitation of a variety of manipulation skills within minutes, using a single demonstration and without any environment instrumentation. It is robust to background clutter and can effectively generalize across environment variations between demonstrator and imitator, greatly outperforming unstructured non-factorized full-frame CNN encodings of previous works [2].

Cite this Paper

BibTeX
@InProceedings{pmlr-v100-sieb20a, title = {Graph-Structured Visual Imitation}, author = {Sieb, Maximilian and Xian, Zhou and Huang, Audrey and Kroemer, Oliver and Fragkiadaki, Katerina}, booktitle = {Proceedings of the Conference on Robot Learning}, pages = {979--989}, year = {2020}, editor = {Kaelbling, Leslie Pack and Kragic, Danica and Sugiura, Komei}, volume = {100}, series = {Proceedings of Machine Learning Research}, month = {30 Oct--01 Nov}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v100/sieb20a/sieb20a.pdf}, url = {https://proceedings.mlr.press/v100/sieb20a.html}, abstract = {We cast visual imitation as a visual correspondence problem. Our robotic agent is rewarded when its actions result in better matching of relative spatial configurations for corresponding visual entities detected in its workspace and the teacher’s demonstration. We build upon recent advances in Computer Vision, such as human finger keypoint detectors, object detectors trained on-the-fly with synthetic augmentations, and point detectors supervised by viewpoint changes [1] and learn multiple visual entity detectors for each demonstration without human annotations or robot interactions. We empirically show that the proposed factorized visual representations of entities and their spatial arrangements drive successful imitation of a variety of manipulation skills within minutes, using a single demonstration and without any environment instrumentation. It is robust to background clutter and can effectively generalize across environment variations between demonstrator and imitator, greatly outperforming unstructured non-factorized full-frame CNN encodings of previous works [2].} }
Endnote
%0 Conference Paper %T Graph-Structured Visual Imitation %A Maximilian Sieb %A Zhou Xian %A Audrey Huang %A Oliver Kroemer %A Katerina Fragkiadaki %B Proceedings of the Conference on Robot Learning %C Proceedings of Machine Learning Research %D 2020 %E Leslie Pack Kaelbling %E Danica Kragic %E Komei Sugiura %F pmlr-v100-sieb20a %I PMLR %P 979--989 %U https://proceedings.mlr.press/v100/sieb20a.html %V 100 %X We cast visual imitation as a visual correspondence problem. Our robotic agent is rewarded when its actions result in better matching of relative spatial configurations for corresponding visual entities detected in its workspace and the teacher’s demonstration. We build upon recent advances in Computer Vision, such as human finger keypoint detectors, object detectors trained on-the-fly with synthetic augmentations, and point detectors supervised by viewpoint changes [1] and learn multiple visual entity detectors for each demonstration without human annotations or robot interactions. We empirically show that the proposed factorized visual representations of entities and their spatial arrangements drive successful imitation of a variety of manipulation skills within minutes, using a single demonstration and without any environment instrumentation. It is robust to background clutter and can effectively generalize across environment variations between demonstrator and imitator, greatly outperforming unstructured non-factorized full-frame CNN encodings of previous works [2].
APA
Sieb, M., Xian, Z., Huang, A., Kroemer, O. & Fragkiadaki, K.. (2020). Graph-Structured Visual Imitation. Proceedings of the Conference on Robot Learning, in Proceedings of Machine Learning Research 100:979-989 Available from https://proceedings.mlr.press/v100/sieb20a.html.

Related Material