Learning from Suboptimal Demonstration via Self-Supervised Reward Regression

Letian Chen, Rohan Paleja, Matthew Gombolay
Proceedings of the 2020 Conference on Robot Learning, PMLR 155:1262-1277, 2021.

Abstract

Learning from Demonstration (LfD) seeks to democratize robotics by enabling non-roboticist end-users to teach robots to perform a task by providing a human demonstration. However, modern LfD techniques, e.g. inverse reinforcement learning (IRL), assume users provide at least stochastically optimal demonstrations. This assumption fails to hold in most real-world scenarios. Recent attempts to learn from sub-optimal demonstration leverage pairwise rankings and following the Luce-Shepard rule. However, we show these approaches make incorrect assumptions and thus suffer from brittle, degraded performance. We overcome these limitations in developing a novel approach that bootstraps off suboptimal demonstrations to synthesize optimality-parameterized data to train an idealized reward function. We empirically validate we learn an idealized reward function with  0.95 correlation with ground-truth reward versus  0.75 for prior work. We can then train policies achieving  200% improvement over the suboptimal demonstration and  90% improvement over prior work. We present a physical demonstration of teaching a robot a topspin strike in table tennis that achieves 32% faster returns and 40% more topspin than user demonstration.

Cite this Paper

BibTeX
@InProceedings{pmlr-v155-chen21b, title = {Learning from Suboptimal Demonstration via Self-Supervised Reward Regression}, author = {Chen, Letian and Paleja, Rohan and Gombolay, Matthew}, booktitle = {Proceedings of the 2020 Conference on Robot Learning}, pages = {1262--1277}, 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/chen21b/chen21b.pdf}, url = {https://proceedings.mlr.press/v155/chen21b.html}, abstract = {Learning from Demonstration (LfD) seeks to democratize robotics by enabling non-roboticist end-users to teach robots to perform a task by providing a human demonstration. However, modern LfD techniques, e.g. inverse reinforcement learning (IRL), assume users provide at least stochastically optimal demonstrations. This assumption fails to hold in most real-world scenarios. Recent attempts to learn from sub-optimal demonstration leverage pairwise rankings and following the Luce-Shepard rule. However, we show these approaches make incorrect assumptions and thus suffer from brittle, degraded performance. We overcome these limitations in developing a novel approach that bootstraps off suboptimal demonstrations to synthesize optimality-parameterized data to train an idealized reward function. We empirically validate we learn an idealized reward function with  0.95 correlation with ground-truth reward versus  0.75 for prior work. We can then train policies achieving  200% improvement over the suboptimal demonstration and  90% improvement over prior work. We present a physical demonstration of teaching a robot a topspin strike in table tennis that achieves 32% faster returns and 40% more topspin than user demonstration.} }
Endnote
%0 Conference Paper %T Learning from Suboptimal Demonstration via Self-Supervised Reward Regression %A Letian Chen %A Rohan Paleja %A Matthew Gombolay %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-chen21b %I PMLR %P 1262--1277 %U https://proceedings.mlr.press/v155/chen21b.html %V 155 %X Learning from Demonstration (LfD) seeks to democratize robotics by enabling non-roboticist end-users to teach robots to perform a task by providing a human demonstration. However, modern LfD techniques, e.g. inverse reinforcement learning (IRL), assume users provide at least stochastically optimal demonstrations. This assumption fails to hold in most real-world scenarios. Recent attempts to learn from sub-optimal demonstration leverage pairwise rankings and following the Luce-Shepard rule. However, we show these approaches make incorrect assumptions and thus suffer from brittle, degraded performance. We overcome these limitations in developing a novel approach that bootstraps off suboptimal demonstrations to synthesize optimality-parameterized data to train an idealized reward function. We empirically validate we learn an idealized reward function with  0.95 correlation with ground-truth reward versus  0.75 for prior work. We can then train policies achieving  200% improvement over the suboptimal demonstration and  90% improvement over prior work. We present a physical demonstration of teaching a robot a topspin strike in table tennis that achieves 32% faster returns and 40% more topspin than user demonstration.
APA
Chen, L., Paleja, R. & Gombolay, M.. (2021). Learning from Suboptimal Demonstration via Self-Supervised Reward Regression. Proceedings of the 2020 Conference on Robot Learning, in Proceedings of Machine Learning Research 155:1262-1277 Available from https://proceedings.mlr.press/v155/chen21b.html.

Related Material