English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Offline reinforcement learning (RL) refers to the problem of learningpolicies from a static dataset of environment interactions. Offline RL enablesextensive use and re-use of historical datasets, while also alleviating safetyconcerns associated with online exploration, thereby expanding the real-worldapplicability of RL. Most prior work in offline RL has focused on tasks withcompact state representations. However, the ability to learn directly from richobservation spaces like images is critical for real-world applications such asrobotics. In this work, we build on recent advances in model-based algorithmsfor offline RL, and extend them to high-dimensional visual observation spaces.Model-based offline RL algorithms have achieved state of the art results instate based tasks and have strong theoretical guarantees. However, they relycrucially on the ability to quantify uncertainty in the model predictions,which is particularly challenging with image observations. To overcome thischallenge, we propose to learn a latent-state dynamics model, and represent theuncertainty in the latent space. Our approach is both tractable in practice andcorresponds to maximizing a lower bound of the ELBO in the unknown POMDP. Inexperiments on a range of challenging image-based locomotion and manipulationtasks, we find that our algorithm significantly outperforms previous offlinemodel-free RL methods as well as state-of-the-art online visual model-based RLmethods. Moreover, we also find that our approach excels on an image-baseddrawer closing task on a real robot using a pre-existing dataset. All resultsincluding videos can be found online at https://sites.google.com/view/lompo/ .

Offline Reinforcement Learning from Images with Latent Space Models