Bidirectional Model-based Policy Optimization

Hang Lai, Jian Shen, Weinan Zhang, Yong Yu
Proceedings of the 37th International Conference on Machine Learning, PMLR 119:5618-5627, 2020.

Abstract

Model-based reinforcement learning approaches leverage a forward dynamics model to support planning and decision making, which, however, may fail catastrophically if the model is inaccurate. Although there are several existing methods dedicated to combating the model error, the potential of the single forward model is still limited. In this paper, we propose to additionally construct a backward dynamics model to reduce the reliance on accuracy in forward model predictions. We develop a novel method, called Bidirectional Model-based Policy Optimization (BMPO) to utilize both the forward model and backward model to generate short branched rollouts for policy optimization. Furthermore, we theoretically derive a tighter bound of return discrepancy, which shows the superiority of BMPO against the one using merely the forward model. Extensive experiments demonstrate that BMPO outperforms state-of-the-art model-based methods in terms of sample efficiency and asymptotic performance.

Cite this Paper


BibTeX
@InProceedings{pmlr-v119-lai20b, title = {Bidirectional Model-based Policy Optimization}, author = {Lai, Hang and Shen, Jian and Zhang, Weinan and Yu, Yong}, booktitle = {Proceedings of the 37th International Conference on Machine Learning}, pages = {5618--5627}, year = {2020}, editor = {Hal Daumé III and Aarti Singh}, volume = {119}, series = {Proceedings of Machine Learning Research}, month = {13--18 Jul}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v119/lai20b/lai20b.pdf}, url = { http://proceedings.mlr.press/v119/lai20b.html }, abstract = {Model-based reinforcement learning approaches leverage a forward dynamics model to support planning and decision making, which, however, may fail catastrophically if the model is inaccurate. Although there are several existing methods dedicated to combating the model error, the potential of the single forward model is still limited. In this paper, we propose to additionally construct a backward dynamics model to reduce the reliance on accuracy in forward model predictions. We develop a novel method, called Bidirectional Model-based Policy Optimization (BMPO) to utilize both the forward model and backward model to generate short branched rollouts for policy optimization. Furthermore, we theoretically derive a tighter bound of return discrepancy, which shows the superiority of BMPO against the one using merely the forward model. Extensive experiments demonstrate that BMPO outperforms state-of-the-art model-based methods in terms of sample efficiency and asymptotic performance.} }
Endnote
%0 Conference Paper %T Bidirectional Model-based Policy Optimization %A Hang Lai %A Jian Shen %A Weinan Zhang %A Yong Yu %B Proceedings of the 37th International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2020 %E Hal Daumé III %E Aarti Singh %F pmlr-v119-lai20b %I PMLR %P 5618--5627 %U http://proceedings.mlr.press/v119/lai20b.html %V 119 %X Model-based reinforcement learning approaches leverage a forward dynamics model to support planning and decision making, which, however, may fail catastrophically if the model is inaccurate. Although there are several existing methods dedicated to combating the model error, the potential of the single forward model is still limited. In this paper, we propose to additionally construct a backward dynamics model to reduce the reliance on accuracy in forward model predictions. We develop a novel method, called Bidirectional Model-based Policy Optimization (BMPO) to utilize both the forward model and backward model to generate short branched rollouts for policy optimization. Furthermore, we theoretically derive a tighter bound of return discrepancy, which shows the superiority of BMPO against the one using merely the forward model. Extensive experiments demonstrate that BMPO outperforms state-of-the-art model-based methods in terms of sample efficiency and asymptotic performance.
APA
Lai, H., Shen, J., Zhang, W. & Yu, Y.. (2020). Bidirectional Model-based Policy Optimization. Proceedings of the 37th International Conference on Machine Learning, in Proceedings of Machine Learning Research 119:5618-5627 Available from http://proceedings.mlr.press/v119/lai20b.html .

Related Material