To Follow or not to Follow: Selective Imitation Learning from Observations

Youngwoon Lee, Edward S. Hu, Zhengyu Yang, Joseph J. Lim
Proceedings of the Conference on Robot Learning, PMLR 100:11-23, 2020.

Abstract

Learning from demonstrations is a useful way to transfer a skill from one agent to another. While most imitation learning methods aim to mimic an expert skill by following the demonstration step-by-step, imitating every step in the demonstration often becomes infeasible when the learner and its environment are different from the demonstration. In this paper, we propose a method that can imitate a demonstration composed solely of observations, which may not be reproducible with the current agent. Our method, dubbed selective imitation learning from observations (SILO), selects reachable states in the demonstration and learns how to reach the selected states. Our experiments on both simulated and real robot environments show that our method reliably performs a new task by following a demonstration. Videos and code are available at https://clvrai.com/silo.

Cite this Paper

BibTeX
@InProceedings{pmlr-v100-lee20a, title = {To Follow or not to Follow: Selective Imitation Learning from Observations}, author = {Lee, Youngwoon and Hu, Edward S. and Yang, Zhengyu and Lim, Joseph J.}, booktitle = {Proceedings of the Conference on Robot Learning}, pages = {11--23}, 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/lee20a/lee20a.pdf}, url = {https://proceedings.mlr.press/v100/lee20a.html}, abstract = {Learning from demonstrations is a useful way to transfer a skill from one agent to another. While most imitation learning methods aim to mimic an expert skill by following the demonstration step-by-step, imitating every step in the demonstration often becomes infeasible when the learner and its environment are different from the demonstration. In this paper, we propose a method that can imitate a demonstration composed solely of observations, which may not be reproducible with the current agent. Our method, dubbed selective imitation learning from observations (SILO), selects reachable states in the demonstration and learns how to reach the selected states. Our experiments on both simulated and real robot environments show that our method reliably performs a new task by following a demonstration. Videos and code are available at https://clvrai.com/silo.} }
Endnote
%0 Conference Paper %T To Follow or not to Follow: Selective Imitation Learning from Observations %A Youngwoon Lee %A Edward S. Hu %A Zhengyu Yang %A Joseph J. Lim %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-lee20a %I PMLR %P 11--23 %U https://proceedings.mlr.press/v100/lee20a.html %V 100 %X Learning from demonstrations is a useful way to transfer a skill from one agent to another. While most imitation learning methods aim to mimic an expert skill by following the demonstration step-by-step, imitating every step in the demonstration often becomes infeasible when the learner and its environment are different from the demonstration. In this paper, we propose a method that can imitate a demonstration composed solely of observations, which may not be reproducible with the current agent. Our method, dubbed selective imitation learning from observations (SILO), selects reachable states in the demonstration and learns how to reach the selected states. Our experiments on both simulated and real robot environments show that our method reliably performs a new task by following a demonstration. Videos and code are available at https://clvrai.com/silo.
APA
Lee, Y., Hu, E.S., Yang, Z. & Lim, J.J.. (2020). To Follow or not to Follow: Selective Imitation Learning from Observations. Proceedings of the Conference on Robot Learning, in Proceedings of Machine Learning Research 100:11-23 Available from https://proceedings.mlr.press/v100/lee20a.html.

Related Material