A Physics-Based Model Prior for Object-Oriented MDPs

Jonathan Scholz, Martin Levihn, Charles Isbell, David Wingate
Proceedings of the 31st International Conference on Machine Learning, PMLR 32(2):1089-1097, 2014.

Abstract

One of the key challenges in using reinforcement learning in robotics is the need for models that capture natural world structure. There are, methods that formalize multi-object dynamics using relational representations, but these methods are not sufficiently compact for real-world robotics. We present a physics-based approach that exploits modern simulation tools to efficiently parameterize physical dynamics. Our results show that this representation can result in much faster learning, by virtue of its strong but appropriate inductive bias in physical environments.

Cite this Paper

BibTeX
@InProceedings{pmlr-v32-scholz14, title = {A Physics-Based Model Prior for Object-Oriented MDPs}, author = {Scholz, Jonathan and Levihn, Martin and Isbell, Charles and Wingate, David}, booktitle = {Proceedings of the 31st International Conference on Machine Learning}, pages = {1089--1097}, year = {2014}, editor = {Xing, Eric P. and Jebara, Tony}, volume = {32}, number = {2}, series = {Proceedings of Machine Learning Research}, address = {Bejing, China}, month = {22--24 Jun}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v32/scholz14.pdf}, url = {https://proceedings.mlr.press/v32/scholz14.html}, abstract = {One of the key challenges in using reinforcement learning in robotics is the need for models that capture natural world structure. There are, methods that formalize multi-object dynamics using relational representations, but these methods are not sufficiently compact for real-world robotics. We present a physics-based approach that exploits modern simulation tools to efficiently parameterize physical dynamics. Our results show that this representation can result in much faster learning, by virtue of its strong but appropriate inductive bias in physical environments.} }
Endnote
%0 Conference Paper %T A Physics-Based Model Prior for Object-Oriented MDPs %A Jonathan Scholz %A Martin Levihn %A Charles Isbell %A David Wingate %B Proceedings of the 31st International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2014 %E Eric P. Xing %E Tony Jebara %F pmlr-v32-scholz14 %I PMLR %P 1089--1097 %U https://proceedings.mlr.press/v32/scholz14.html %V 32 %N 2 %X One of the key challenges in using reinforcement learning in robotics is the need for models that capture natural world structure. There are, methods that formalize multi-object dynamics using relational representations, but these methods are not sufficiently compact for real-world robotics. We present a physics-based approach that exploits modern simulation tools to efficiently parameterize physical dynamics. Our results show that this representation can result in much faster learning, by virtue of its strong but appropriate inductive bias in physical environments.
RIS
TY - CPAPER TI - A Physics-Based Model Prior for Object-Oriented MDPs AU - Jonathan Scholz AU - Martin Levihn AU - Charles Isbell AU - David Wingate BT - Proceedings of the 31st International Conference on Machine Learning DA - 2014/06/18 ED - Eric P. Xing ED - Tony Jebara ID - pmlr-v32-scholz14 PB - PMLR DP - Proceedings of Machine Learning Research VL - 32 IS - 2 SP - 1089 EP - 1097 L1 - http://proceedings.mlr.press/v32/scholz14.pdf UR - https://proceedings.mlr.press/v32/scholz14.html AB - One of the key challenges in using reinforcement learning in robotics is the need for models that capture natural world structure. There are, methods that formalize multi-object dynamics using relational representations, but these methods are not sufficiently compact for real-world robotics. We present a physics-based approach that exploits modern simulation tools to efficiently parameterize physical dynamics. Our results show that this representation can result in much faster learning, by virtue of its strong but appropriate inductive bias in physical environments. ER -
APA
Scholz, J., Levihn, M., Isbell, C. & Wingate, D.. (2014). A Physics-Based Model Prior for Object-Oriented MDPs. Proceedings of the 31st International Conference on Machine Learning, in Proceedings of Machine Learning Research 32(2):1089-1097 Available from https://proceedings.mlr.press/v32/scholz14.html.

Related Material