Despite recent advances in the field of Offline Reinforcement Learning (RL), less attention has been paid to understanding the behaviors of learned RL agents. As a result, there remain some gaps in our understandings, i.e., why is one offline RL agent more performant than another? In this work, we first introduce a set of experiments to evaluate offline RL agents, focusing on three fundamental aspects: representations, value functions and policies. Counterintuitively, we show that a more performant offline RL agent can learn relatively low-quality representations and inaccurate value functions. Furthermore, we showcase that the proposed experiment setups can be effectively used to diagnose the bottleneck of offline RL agents. Inspired by the evaluation results, a novel offline RL algorithm is proposed by a simple modification of IQL and achieves SOTA performance. Finally, we investigate when a learned dynamics model is helpful to model-free offline RL agents, and introduce an uncertainty-based sample selection method to mitigate the problem of model noises. Code is available at: https://github.com/fuyw/RIQL.

Despite recent advances in the field of Offline Reinforcement Learning (RL), less attention has been paid to understanding the behaviors of learned RL agents. As a result, there remain some gaps in our understandings, i.e., why is one offline RL agent more performant than another? In this work, we first introduce a set of experiments to evaluate offline RL agents, focusing on three fundamental aspects: representations, value functions and policies. Counterintuitively, we show that a more performant offline RL agent can learn relatively low-quality representations and inaccurate value functions. Furthermore, we showcase that the proposed experiment setups can be effectively used to diagnose the bottleneck of offline RL agents.

Offline reinforcement learning (RL) aims to learn an effective policy from a pre-collected dataset. Most existing works are to develop sophisticated learning algorithms, with less emphasis on improving the data collection process. Moreover, it is even challenging to extend the single-task setting and collect a task-agnostic dataset that allows an agent to perform multiple downstream tasks. In this paper, we propose a Curiosity-driven Unsupervised Data Collection (CUDC) method to expand feature space using adaptive temporal distances for task-agnostic data collection and ultimately improve learning efficiency and capabilities for multi-task offline RL. To achieve this, CUDC estimates the probability of the k-step future states being reachable from the current states, and adapts how many steps into the future that the dynamics model should predict. With this adaptive reachability mechanism in place, the feature representation can be diversified, and the agent can navigate itself to collect higher-quality data with curiosity. Empirically, CUDC surpasses existing unsupervised methods in efficiency and learning performance in various downstream offline RL tasks of the DeepMind control suite.