Challenges and Opportunities in Offline Reinforcement Learning from Visual Observations

Offline reinforcement learning has shown great promise in leveraging large pre-collected datasets for policy learning, allowing agents to forgo often-expensive online data collection. However, to date, offline reinforcement learning from visual observations has been relatively under-explored, and there is a lack of understanding of where the remaining challenges lie. In this paper, we seek to establish simple baselines for continuous control in the visual domain. We show that simple modifications to two state-of-the-art vision-based online reinforcement learning algorithms, DreamerV2 and DrQ-v2, suffice to outperform prior work and establish a competitive baseline. We rigorously evaluate these algorithms on both existing offline datasets and a new testbed for offline reinforcement learning from visual observations that better represents the data distributions present in real-world offline reinforcement learning problems, and open-source our code and data to facilitate progress in this important domain. Finally, we present and analyze several key desiderata unique to offline RL from visual observations, including visual distractions and visually identifiable changes in dynamics. [19, 20], and predicts ahead using compact model latent states. The particular instantiation used in DreamerV2 uses model states s t containing a deterministic component h t , implemented as the recurrent state of a Gated Recurrent Unit (GRU, [9]), and a stochastic component z t with categorical distribution. The actor and critic are trained from imagined trajectories of latent states, starting at encoded states of previously encountered sequences.