Adding a reference in the README to new benchmarking results for standard offline RL tasks

[TheotimeLH]

Hello @seohongpark ,

Thanks a lot for creating this very useful benchmark!

In our recent work Guided flow policy (GFP), we conducted an extensive evaluation on OGBench standard offline RL tasks ("singletask"). We evaluated GFP on 105 tasks from OGBench, and when needed, we also evaluated ReBRAC and FQL. Including antmaze-large-navigate/stitch/explore, antmaze-giant-navigate, humanoidmaze-medium-navigate/stich, humanoidmaze-large-navigate, antsoccer-arena-navigate/stitch, cube-single/double/triple-play/noisy, puzzle-4x4-play/noisy, scene-play/noisy and some visual tasks.

To our knowledge, some of these tasks had not been evaluated before. We see this as our contribution to support OGBench, which we consider the most comprehensive benchmark in Offline RL. For ease of use, we included CSV files of the exact results we got (in this folder).

Given the scope of our evaluation, would you consider adding a reference to GFP's repo in OGBench's README, alongside FQL?
At the line

For standard offline RL, we do not provide official reference implementations or benchmarking results. However, you may find implementations of some offline RL algorithms (e.g., IQL, ReBRAC, and FQL) with partial benchmarking results in this repository.

We believe this would be beneficial for the community, as it provides additional benchmarking on OGBench, and it would mean a lot to us.

Best,

Theotime