Deployment-Efficient Reinforcement Learning via Model-Based Offline Optimization

Abstract

Most reinforcement learning (RL) algorithms assume online access to the environment, in which one may readily interleave updates to the policy with experience collection using that policy. However, in many real-world applications such as health, education, dialogue agents, and robotics, the cost or potential risk of deploying a new data-collection policy is high, to the point that it can become prohibitive to update the data-collection policy more than a few times during learning. With this view, we propose a novel concept of deployment efficiency, measuring the number of distinct data-collection policies that are used during policy learning. We observe that naïvely applying existing model-free offline RL algorithms recursively does not lead to a practical deployment-efficient and sample-efficient algorithm. We propose a novel model-based algorithm, Behavior-Regularized Model-ENsemble (BREMEN) that can effectively optimize a policy offline using 10-20 times fewer data than prior works. Furthermore, the recursive application of BREMEN is able to achieve impressive deployment efficiency while maintaining the same or better sample efficiency, learning successful policies from scratch on simulated robotic environments with only 5-10 deployments, compared to typical values of hundreds to millions in standard RL baselines. Codes and pre-trained models are available at https://github.com/matsuolab/BREMEN. * Equal contribution.Preprint. Under review.

A Minimalist Approach to Offline Reinforcement Learning

“…Figure 3 shows instability over the set of evaluations, which means the performance of the agent may be dependent on the specific stopping point chosen for evaluation. This questions the empirical effectiveness of offline RL for safety-critical real-world use cases [Mandel et al, 2014, Gottesman et al, 2018, Gauci et al, 2018, Jaques et al, 2019, Matsushima et al, 2020 as well as the current trend of reporting only the mean-value of the final policy in offline benchmarking [Fu et al, 2021].…”

Autoregressive Dynamics Models for Offline Policy Evaluation and Optimization

“…While the use of model-based principles in OPE has been relatively rare, it has been more commonly used for policy optimization. The field of model-based RL has matured in recent years to yield impressive results for both online (Nagabandi et al, 2018;Chua et al, 2018;Kurutach et al, 2018;Janner et al, 2019) and offline (Matsushima et al, 2020;Kidambi et al, 2020;Yu et al, 2020;Argenson and Dulac-Arnold, 2020) policy optimization. Several of the techniques we employ, such For both family of dynamics models, we train 48 models with different hyperparameters.…”

Discovering an Aid Policy to Minimize Student Evasion Using Offline Reinforcement Learning

“…Deployment efficiency measure [22] in RL counts the number of changes in the data-collection policy during learning, i.e., an offline RL setup corresponds to a single deployment allowed for learning. Reinforcement learning methods can be classified through data and interaction perspectives [23].…”

“…Recently, fully offline reinforcement learning methods like Random Ensemble Mixture (REM) [12], Deep Q-learning from Demonstrations (DQfD) [24], Bootstrapping Error Accumulation Reduction (BEAR) [25], Batch-Constrained deep Q-learning (BCQ) [13], [26], and Behavior Regularized Actor Critic (BRAC) [27] consider different approaches and techniques to surpass those limitations. Behavior-Regularized Model-ENsemble (BREMEN) [22] is a model-based algorithm that can be used in a fully offline setup, but its main goal is to be efficient considering the number of changes in the data-collection policy needed during learning (deployment efficiency) and sampling efficient by using a mixed (online and offline) approach. Similar to BREMEN, Advantage Weighted Actor Critic (AWAC) [28] is focused on effectively fine-tuning using online experiences after an offline pre-training period.…”