Learning Variational Latent Dynamics: Towards Model-based Imitation and Control. Yin, H., Melo, F. S., Billard, A., & Paiva, A. Technical Report GAIPS-TR-001-17, Intelligent Agents and Synthetic Characters Group, May, 2018. Paper abstract bibtex In this paper, we learn dynamics from highdimensional demonstrations to facilitate model-based prediction and robot control. The proposed approach leverages the progress in variational-bayes and sequence modeling, extracting a low-dimensional latent space so the dynamical relations of interest can be compactly represented and learned. Different from existing works, our model captures latent dynamics in a more general form and features efficient inference for pattern filtering, prediction and synthesis. The extracted feature mapping and latent dynamics can be naturally integrated in robot learning, yielding task imitation from raw data and prediction-based reproduction. The performance of latent dynamics learning and model-based imitation is shown in three tasks: 1) reconstructing and predicting images of bouncing balls movement with an accuracy competitive to the state-of-the art; 2) synthesizing diverse handwriting image sequences; 3) learning to strike a ball under partial visual input, with results significantly outperforming baselines.
@techreport {Yin2018-ID1018,
abstract = {In this paper, we learn dynamics from highdimensional
demonstrations to facilitate model-based prediction
and robot control. The proposed approach leverages the
progress in variational-bayes and sequence modeling, extracting
a low-dimensional latent space so the dynamical relations of
interest can be compactly represented and learned. Different
from existing works, our model captures latent dynamics
in a more general form and features efficient inference for
pattern filtering, prediction and synthesis. The extracted feature
mapping and latent dynamics can be naturally integrated in
robot learning, yielding task imitation from raw data and
prediction-based reproduction. The performance of latent dynamics
learning and model-based imitation is shown in three
tasks: 1) reconstructing and predicting images of bouncing balls
movement with an accuracy competitive to the state-of-the art;
2) synthesizing diverse handwriting image sequences; 3)
learning to strike a ball under partial visual input, with results
significantly outperforming baselines.},
institution = {Intelligent Agents and Synthetic Characters Group},
month = {May},
number = {GAIPS-TR-001-17},
title = {Learning Variational Latent Dynamics: Towards Model-based Imitation and Control},
year = {2018},
author = {Yin, H. and Melo, F. S. and Billard, A. and Paiva, A.},
url = {{https://github.com/navigator8972/navigator8972.github.io/raw/master/files/learnvaedyn2018.pdf}},
}
Downloads: 0
{"_id":"R8qCM8DWSSuTjZWvL","bibbaseid":"yin-melo-billard-paiva-learningvariationallatentdynamicstowardsmodelbasedimitationandcontrol-2018","authorIDs":["4qDAXT4gdnPdBNGLb","5da354f7d96d95db01000023","5dea364d07a0c8df01000060","5e16a01eda4561de01000091","5e1cb3d37723aadf01000128","5e2ada16638921df0100012d","5e69a82f23ebccde010000d5","6Qq6mgezs4zZxncC5","8znxuZQuncFQi59Lj","CYPZuY3RK9D5Pc65s","EzGtfdDdhDoqnfZcX","KGybfkfDE4enFkaa5","XfXsNHL2xdfac3iiz","danpWtv6sR6ErcP9y","h75RY3NTCod4WceQa","hygdvDkWHWwLzzxTM","jzXk96RY4YfLh3jGA","mSSsYDAqZ8nPyfL8t","pwZANZmrmp9vd2nX5","sFT62jXGfPtZFbE33","uucdkJx5gZWNAkJgT"],"author_short":["Yin, H.","Melo, F. S.","Billard, A.","Paiva, A."],"bibdata":{"bibtype":"techreport","type":"techreport","abstract":"In this paper, we learn dynamics from highdimensional demonstrations to facilitate model-based prediction and robot control. The proposed approach leverages the progress in variational-bayes and sequence modeling, extracting a low-dimensional latent space so the dynamical relations of interest can be compactly represented and learned. Different from existing works, our model captures latent dynamics in a more general form and features efficient inference for pattern filtering, prediction and synthesis. The extracted feature mapping and latent dynamics can be naturally integrated in robot learning, yielding task imitation from raw data and prediction-based reproduction. The performance of latent dynamics learning and model-based imitation is shown in three tasks: 1) reconstructing and predicting images of bouncing balls movement with an accuracy competitive to the state-of-the art; 2) synthesizing diverse handwriting image sequences; 3) learning to strike a ball under partial visual input, with results significantly outperforming baselines.","institution":"Intelligent Agents and Synthetic Characters Group","month":"May","number":"GAIPS-TR-001-17","title":"Learning Variational Latent Dynamics: Towards Model-based Imitation and Control","year":"2018","author":[{"propositions":[],"lastnames":["Yin"],"firstnames":["H."],"suffixes":[]},{"propositions":[],"lastnames":["Melo"],"firstnames":["F.","S."],"suffixes":[]},{"propositions":[],"lastnames":["Billard"],"firstnames":["A."],"suffixes":[]},{"propositions":[],"lastnames":["Paiva"],"firstnames":["A."],"suffixes":[]}],"url":"https://github.com/navigator8972/navigator8972.github.io/raw/master/files/learnvaedyn2018.pdf","bibtex":"@techreport {Yin2018-ID1018,\n\tabstract = {In this paper, we learn dynamics from highdimensional\ndemonstrations to facilitate model-based prediction\nand robot control. The proposed approach leverages the\nprogress in variational-bayes and sequence modeling, extracting\na low-dimensional latent space so the dynamical relations of\ninterest can be compactly represented and learned. Different\nfrom existing works, our model captures latent dynamics\nin a more general form and features efficient inference for\npattern filtering, prediction and synthesis. The extracted feature\nmapping and latent dynamics can be naturally integrated in\nrobot learning, yielding task imitation from raw data and\nprediction-based reproduction. The performance of latent dynamics\nlearning and model-based imitation is shown in three\ntasks: 1) reconstructing and predicting images of bouncing balls\nmovement with an accuracy competitive to the state-of-the art;\n2) synthesizing diverse handwriting image sequences; 3)\nlearning to strike a ball under partial visual input, with results\nsignificantly outperforming baselines.},\n\tinstitution = {Intelligent Agents and Synthetic Characters Group},\n\tmonth = {May},\n\tnumber = {GAIPS-TR-001-17},\n\ttitle = {Learning Variational Latent Dynamics: Towards Model-based Imitation and Control},\n\tyear = {2018},\n\tauthor = {Yin, H. and Melo, F. S. and Billard, A. and Paiva, A.},\n\n url = {{https://github.com/navigator8972/navigator8972.github.io/raw/master/files/learnvaedyn2018.pdf}},\n}\n\n","author_short":["Yin, H.","Melo, F. S.","Billard, A.","Paiva, A."],"key":"Yin2018-ID1018","id":"Yin2018-ID1018","bibbaseid":"yin-melo-billard-paiva-learningvariationallatentdynamicstowardsmodelbasedimitationandcontrol-2018","role":"author","urls":{"Paper":"https://github.com/navigator8972/navigator8972.github.io/raw/master/files/learnvaedyn2018.pdf"},"metadata":{"authorlinks":{"yin, h":"https://navigator8972.github.io/navigator8972.github.io/publications/"}}},"bibtype":"techreport","biburl":"https://raw.githubusercontent.com/navigator8972/navigator8972.github.io/master/files/mypublications.bib","creationDate":"2019-10-13T16:49:13.097Z","downloads":0,"keywords":[],"search_terms":["learning","variational","latent","dynamics","towards","model","based","imitation","control","yin","melo","billard","paiva"],"title":"Learning Variational Latent Dynamics: Towards Model-based Imitation and Control","year":2018,"dataSources":["3DWEKzfgRRQEyNBkf"]}