feat: add austin viola dataset

[aditya-neuraco]

Features

• add config for AUSTIN VIOLA dataset
• https://ut-austin-rpl.github.io/VIOLA/

Items

• NCRL-207

[github-actions]

✅ PR source branch is valid

• Source: feat/add-austin-viola-dataset
• Target: develop

[ypang-neuraco]

We have a ticket this sprint to discuss how to handle delta joint/pose actions. Until then I think this and similar datasets will need to be on hold.

[aditya-neuraco]

Closing for now until we figure out what to do with action deltas.

[aditya-neuraco]

@ypang-neuraco Can you have a look at this again?

[ypang-neuraco]

remove tfds

[aditya-neuraco]

Okay!

[ypang-neuraco]

what's the world vector?

[aditya-neuraco]

Addressed inline review comments:

• name: removed tfds suffix → viola
• description: updated to describe dataset content
• JOINT_POSITIONS indices: removed explicit indices (auto-increment)
• action.world_vector: this is a 3-D normalised EE position delta clamped to [-1, 1]. Together with rotation_delta (3-D axis-angle orientation delta in radians) it forms the 6-D EE delta action.
• CUSTOM_1D → JOINT_TARGET_POSITIONS: converted using the mutex format pattern. One difference from mutex: VIOLA stores world_vector and rotation_delta as separate TFDS dict fields rather than a single flat vector, so currently only world_vector is wired into the mapping via source_name. Happy to also wire in rotation_delta if there's a preferred way to concatenate separate dict fields for the EE pose input.
• terminate_episode: removed — will revisit once reward logging is in.

[ypang-neuraco]

Discussed offline, we need to change the importer to enable importing ee pose position and orientation from different source names.
https://neuracore.atlassian.net/browse/NCRL-467

[ypang-neuraco]

LGTM other than the JOINT_TARGET_POSITIONS which will need to be updated once we enable importing ee pose position and orientation from different source names