Offline Reinforcement Learning: Tutorial, Review, and Perspectives on Open Problems
By Sergey Levine, Aviral Kumar, George Tucker, Justin Fu. arXiv.org, 2020
2,718 citations recorded by the source corpus.
In this tutorial article, we aim to provide the reader with the conceptual tools needed to get started on research on offline reinforcement learning algorithms: reinforcement learning algorithms that utilize previously collected data, without additional online data collection. Offline reinforcement learning algorithms hold tremendous promise for making it possible to turn large datasets into powerful decision making engines. Effective offline reinforcement learning methods would be able to extract policies with the maximum possible utility out of the available data, thereby allowing automation of a wide range of decision-making domains, from healthcare and education to robotics. However, the limitations of current algorithms make this difficult. We will aim to provide the reader with an understanding of these challenges, particularly in the context of modern deep reinforcement learning methods, and describe some potential solutions that have been explored in recent work to mitigate these challenges, along with recent applications, and a discussion of perspectives on open problems in the field.
Subfields: Artificial Intelligence, Industrial and Manufacturing Engineering, Reinforcement Learning
View Paper: https://arxiv.org/abs/2005.01643
How researchers see this paper
The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.
In Artificial Intelligence
- Sergey Levine
- Yoshua Bengio
- Pieter Abbeel
- Ruslan Salakhutdinov
- Mohit Bansal
- Chelsea Finn
- Quoc V. Le
- Jianfeng Gao
In Industrial and Manufacturing Engineering
- Shixiang Shane Gu
- Sergey Levine
- Ofir Nachum
- Chelsea Finn
- Scott Niekum
- Scott Fujimoto
- Aviral Kumar
- Aravind Rajeswaran
In Reinforcement Learning
- Sergey Levine
- Pieter Abbeel
- Chelsea Finn
- Ofir Nachum
- Shixiang Shane Gu
- Aviral Kumar
- Ilya Kostrikov
- George Tucker
Citations in the corpus
- Augmented Language Models: a Survey (2023, methodology): “…such as: the suboptimality of the initial behavior policy, the dimensionality of the state and the action space, the length of the effective horizon, the accumulation of errors and distributional shifts due to the discrepancy between the behavior policy and the learned one (Levine et al., 2020).”
- Decision Transformer: Reinforcement Learning via Sequence Modeling (2021, background): “This task is traditionally challenging due to error propagation and value overestimation (Levine et al., 2020).”
- Toward Causal Representation Learning (2021, background): “Off-line RL [73], [150] tries to address this concern by learning a policy from a fixed data set of trajectories, without requiring any experimental or interventional data (i.e., without any interaction with the environment).”
- Conservative Q-Learning for Offline Reinforcement Learning (2020, methodology): “Most of these methods require a separately estimated model to the behavior policy, πβ(a|s) [15, 30, 59, 27, 54, 55], and are thus limited by their ability to accurately estimate the unknown behavior policy [42], which might be especially complex in settings where the data is collected from multiple ”
- What Matters in Learning from Offline Human Demonstrations for Robot Manipulation (2021, methodology): “By contrast, Batch (Offline) Reinforcement Learning [11, 12] is a method to learn from demonstrations that can consist of both good and bad quality data, by leveraging reward annotations in the datasets.”
- Accelerating Online Reinforcement Learning with Offline Datasets (2020, neutral): “Offline RL methods aim to address this issue, often by constraining the actor updates to avoid excessive deviation from the data distribution [23, 41, 13, 14, 12, 21, 8, 22, 38, 26, 47].”
- Towards Causal Representation Learning (2021, background): “Offline RL [72, 150] tries to address this concern by learning a policy from a fixed dataset of trajectories, without requiring any experimental or interventional data (i.”
- Towards Continual Reinforcement Learning: A Review and Perspectives (2020, background): “While they have been quite successful relative to other approaches to date, replay based strategies generally struggle to effectively leverage past data when the behavior of the current policy is significantly different, as off-policy learning tends to be difficult in this setting (see (Levine et al”
- MOReL : Model-Based Offline Reinforcement Learning (2020, neutral): “[44] for an overview of potential applications.”
- Offline RL Without Off-Policy Evaluation (2021, neutral): “Recent work in offline RL is well summarized by Levine et al. [2020].”
- A Minimalist Approach to Offline Reinforcement Learning (2021, methodology): “, 2019], or policy constraint [Levine et al., 2020]) depending on how this “closeness” is chosen to be implemented.”
- Regularized Behavior Value Estimation (2021, background): “Correspondence to: Caglar Gulcehre <caglarg@google.com>. et al., 2018; Gulcehre et al., 2020; Levine et al., 2020), also known as batch RL (Ernst et al., 2005; Lange et al., 2012), agents learn from a fixed dataset previously logged by other (possibly unknown) agents.”
- Offline Reinforcement Learning with Fisher Divergence Critic Regularization (2021, background): “This setting is known as offline RL and has attracted a significant amount of interest in recent years (Levine et al., 2020; Lange et al., 2012).”
- Continuous Doubly Constrained Batch Reinforcement Learning (2021, background): “Aiming for a practical framework to improve arbitrary existing policies, much research has studied batch RL (Levine et al., 2020; Lange et al., 2012) and the issue of overestimation (Thrun & Schwartz, 1993; Hasselt, 2010; Hasselt et al., 2016).”
- EMaQ: Expected-Max Q-Learning Operator for Simple Yet Effective Offline and Online RL (2020, methodology): “We refer to online RL as the setting where iteratively, a batch of M environment steps with an exploration policy are inter-leaved with M RL updates (Levine et al., 2020; Matsushima et al., 2020).”
- Critic Regularized Regression (2020, background): “For comprehensive in-depth reviews of offline RL, we refer the reader to Lange et al. [19] and Levine et al. [20].”
- Provable Representation Learning for Imitation with Contrastive Fourier Features (2021, background): “The limited availability of target demonstrations stands in contrast to the recent proliferation of large offline datasets for sequential decision making [28, 19, 7, 21].”
- Hyperparameter Selection for Imitation Learning (2021, background): “…recently highlighted the same problem occurring in offline RL (Lagoudakis & Parr, 2003; Ernst et al., 2005; Riedmiller, 2005; Lange et al., 2012; Levine et al., 2020), where HPs are usually selected on the performance of the agent on the online environment (although the core constraint of…”