Motivation

Alongside the D4RL datasets, it could be useful to have a pixel-based continuous control benchmark for offline RL integrated into this codebase. One such existing benchmark is V-D4RL with associated TMLR paper: Challenges and Opportunities in Offline Reinforcement Learning from Visual Observations. This would naturally compliment the low-dimensional D4RL datasets and the proposed discrete action space Atari datasets.

I think this would be appropriate for the"call for contributions" stack. @vmoens

Solution

Integrate V-D4RL datasets from https://github.com/conglu1997/v-d4rl into pytorch/rl

Checklist

• I have checked that there is no similar issue in the repo (required)