Datasets:

---

MINT-SJTU

/

RW-RL-Dataset

like 1

Follow

MINT-SJTU 22

Hi Evo-RL team,

Thanks for open-sourcing the Evo-RL codebase and the initial RW-RL-Dataset samples.

I noticed that the current Hugging Face dataset MINT-SJTU/RW-RL-Dataset seems to contain only a small sample release, while the project README says that the full Hugging Face dataset release is still “coming soon”.

Could you share more details about the future release plan for the full real-world RL dataset?

Specifically, I would like to ask:

1. Will the complete RW-RL-Dataset be released publicly?

2. Is there an expected release timeline?

3. Will the full release include all robot platforms and tasks used in the Evo-RL paper/project?

4. What data fields will be included in the full release? For example:

   • raw multi-view videos
   • robot states
   • actions
   • policy actions
   • human intervention labels
   • success / failure labels
   • value or advantage annotations
   • task instructions and episode metadata

5. Will the dataset be released in LeRobot format?

6. Will preprocessing scripts, training configs, and fixed version tags also be provided for reproducibility?

7. Is the current Hugging Face repo intended only as a sample release, or will it later be expanded into the full dataset?

We are interested in using Evo-RL as a reference pipeline for real-world VLA / WAM training, so the dataset roadmap would be very helpful.

Thanks again for the great work.