MetaWorld version mismatch: v2 symbols required (`ALL_V2_ENVIRONMENTS_GOAL_OBSERVABLE`), PyPI provides v3

[EmbodiedMind]

Summary
I can’t build/run the MetaWorld data generation pipeline in QueST due to a MetaWorld version mismatch. The code expects MetaWorld v2 (now unmaintained), e.g.:

from metaworld.envs import ALL_V2_ENVIRONMENTS_GOAL_OBSERVABLE

but the current MetaWorld on PyPI is v3, so the import fails. As a result, the scripts don’t run out of the box.

---

Minimal Repro

Fresh conda env (Python 3.10), follow QueST install, then:

python scripts/generate_metaworld_dataset.py

Observed error on startup (MetaWorld v3 default from PyPI):

ImportError: cannot import name 'ALL_V2_ENVIRONMENTS_GOAL_OBSERVABLE' from 'metaworld.envs' (.../site-packages/metaworld/envs/__init__.py)

---

Where the v2 API is assumed
The code relies on v2-only registries/constants such as ALL_V2_ENVIRONMENTS_GOAL_OBSERVABLE and v2 env-name patterns like f"{env_name}-goal-observable"—these have changed in v3.

---

Requests / Questions

1. Which exact MetaWorld version does QueST target?

   • If it’s v2, could you please share a constraints/lock file with the exact versions you used for:

     • metaworld, mujoco/mujoco-py, gym/gymnasium, and any other critical pins (e.g., Python, NumPy)?

   • If you’ve moved to v3, could you provide:

     • A mapping for v2 constants and env-name formats (e.g., ALL_V2_ENVIRONMENTS_GOAL_OBSERVABLE, "{env_name}-goal-observable") to the v3 API, or a branch/commit that implements the update?

(Optionally, if helpful, I can provide my exact OS/Python details and re-test any suggested pins/branch.)

---

Why this matters
MetaWorld v2 is unmaintained, while v3 is current but has API changes. A pinned v2 environment (or a v3-compatible path) would make it much easier for new users to reproduce the MetaWorld data generation.

Thanks a lot for the great project—happy to test a v3 branch or verify a pinned v2 environment on a clean machine if you can share the versions you used!