Continuous Deep Q-Learning with Model-based Acceleration

Shixiang Gu, Timothy Lillicrap, Ilya Sutskever, Sergey Levine
Proceedings of The 33rd International Conference on Machine Learning, PMLR 48:2829-2838, 2016.

Abstract

Model-free reinforcement learning has been successfully applied to a range of challenging problems, and has recently been extended to handle large neural network policies and value functions. However, the sample complexity of model-free algorithms, particularly when using high-dimensional function approximators, tends to limit their applicability to physical systems. In this paper, we explore algorithms and representations to reduce the sample complexity of deep reinforcement learning for continuous control tasks. We propose two complementary techniques for improving the efficiency of such algorithms. First, we derive a continuous variant of the Q-learning algorithm, which we call normalized advantage functions (NAF), as an alternative to the more commonly used policy gradient and actor-critic methods. NAF representation allows us to apply Q-learning with experience replay to continuous tasks, and substantially improves performance on a set of simulated robotic control tasks. To further improve the efficiency of our approach, we explore the use of learned models for accelerating model-free reinforcement learning. We show that iteratively refitted local linear models are especially effective for this, and demonstrate substantially faster learning on domains where such models are applicable.

Cite this Paper


BibTeX
@InProceedings{pmlr-v48-gu16, title = {Continuous Deep Q-Learning with Model-based Acceleration}, author = {Gu, Shixiang and Lillicrap, Timothy and Sutskever, Ilya and Levine, Sergey}, booktitle = {Proceedings of The 33rd International Conference on Machine Learning}, pages = {2829--2838}, year = {2016}, editor = {Balcan, Maria Florina and Weinberger, Kilian Q.}, volume = {48}, series = {Proceedings of Machine Learning Research}, address = {New York, New York, USA}, month = {20--22 Jun}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v48/gu16.pdf}, url = { http://proceedings.mlr.press/v48/gu16.html }, abstract = {Model-free reinforcement learning has been successfully applied to a range of challenging problems, and has recently been extended to handle large neural network policies and value functions. However, the sample complexity of model-free algorithms, particularly when using high-dimensional function approximators, tends to limit their applicability to physical systems. In this paper, we explore algorithms and representations to reduce the sample complexity of deep reinforcement learning for continuous control tasks. We propose two complementary techniques for improving the efficiency of such algorithms. First, we derive a continuous variant of the Q-learning algorithm, which we call normalized advantage functions (NAF), as an alternative to the more commonly used policy gradient and actor-critic methods. NAF representation allows us to apply Q-learning with experience replay to continuous tasks, and substantially improves performance on a set of simulated robotic control tasks. To further improve the efficiency of our approach, we explore the use of learned models for accelerating model-free reinforcement learning. We show that iteratively refitted local linear models are especially effective for this, and demonstrate substantially faster learning on domains where such models are applicable.} }
Endnote
%0 Conference Paper %T Continuous Deep Q-Learning with Model-based Acceleration %A Shixiang Gu %A Timothy Lillicrap %A Ilya Sutskever %A Sergey Levine %B Proceedings of The 33rd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2016 %E Maria Florina Balcan %E Kilian Q. Weinberger %F pmlr-v48-gu16 %I PMLR %P 2829--2838 %U http://proceedings.mlr.press/v48/gu16.html %V 48 %X Model-free reinforcement learning has been successfully applied to a range of challenging problems, and has recently been extended to handle large neural network policies and value functions. However, the sample complexity of model-free algorithms, particularly when using high-dimensional function approximators, tends to limit their applicability to physical systems. In this paper, we explore algorithms and representations to reduce the sample complexity of deep reinforcement learning for continuous control tasks. We propose two complementary techniques for improving the efficiency of such algorithms. First, we derive a continuous variant of the Q-learning algorithm, which we call normalized advantage functions (NAF), as an alternative to the more commonly used policy gradient and actor-critic methods. NAF representation allows us to apply Q-learning with experience replay to continuous tasks, and substantially improves performance on a set of simulated robotic control tasks. To further improve the efficiency of our approach, we explore the use of learned models for accelerating model-free reinforcement learning. We show that iteratively refitted local linear models are especially effective for this, and demonstrate substantially faster learning on domains where such models are applicable.
RIS
TY - CPAPER TI - Continuous Deep Q-Learning with Model-based Acceleration AU - Shixiang Gu AU - Timothy Lillicrap AU - Ilya Sutskever AU - Sergey Levine BT - Proceedings of The 33rd International Conference on Machine Learning DA - 2016/06/11 ED - Maria Florina Balcan ED - Kilian Q. Weinberger ID - pmlr-v48-gu16 PB - PMLR DP - Proceedings of Machine Learning Research VL - 48 SP - 2829 EP - 2838 L1 - http://proceedings.mlr.press/v48/gu16.pdf UR - http://proceedings.mlr.press/v48/gu16.html AB - Model-free reinforcement learning has been successfully applied to a range of challenging problems, and has recently been extended to handle large neural network policies and value functions. However, the sample complexity of model-free algorithms, particularly when using high-dimensional function approximators, tends to limit their applicability to physical systems. In this paper, we explore algorithms and representations to reduce the sample complexity of deep reinforcement learning for continuous control tasks. We propose two complementary techniques for improving the efficiency of such algorithms. First, we derive a continuous variant of the Q-learning algorithm, which we call normalized advantage functions (NAF), as an alternative to the more commonly used policy gradient and actor-critic methods. NAF representation allows us to apply Q-learning with experience replay to continuous tasks, and substantially improves performance on a set of simulated robotic control tasks. To further improve the efficiency of our approach, we explore the use of learned models for accelerating model-free reinforcement learning. We show that iteratively refitted local linear models are especially effective for this, and demonstrate substantially faster learning on domains where such models are applicable. ER -
APA
Gu, S., Lillicrap, T., Sutskever, I. & Levine, S.. (2016). Continuous Deep Q-Learning with Model-based Acceleration. Proceedings of The 33rd International Conference on Machine Learning, in Proceedings of Machine Learning Research 48:2829-2838 Available from http://proceedings.mlr.press/v48/gu16.html .

Related Material