[bokalnv]

Summary

Stacked on #90. This PR adds the episode checks and the first validity metrics.

Changes

• Episode checks run before any metric.
  • Correctness checks: finite values, aligned per-step lengths, valid observed rotations, and pose actions within
    their clip bound.
  • Camera checks: one frame per step, no black frames, no frozen runs. Frames are read in chunks.
  • An episode that fails a check is counted in correctness_pass_rate or image_stream_pass_rate and left out of
    every dataset metric. The report lists the failures per check, with examples.
• final_success_rate: the share of episodes whose task success holds at their last step.
• action_rotation_error: how far the commanded target rotations are from valid rotations.
• A metric declared with baseline=True is also computed on the source dataset's valid episodes.
• Docs: the metrics page lists the checks, generated from their @check declarations.

Testing

• Unit tests: 395 passed. One known-answer test per check and per metric.
• All 11 shipped annotated datasets pass every check.
• On generated data:
  • Franka MimicGen, 50 demos: 50/50 episodes valid; final success 0.92.
  • GR1 with cameras: 4/4 episodes valid; image checks pass.
• End-to-end tests: MimicGen and SkillGen bin pass with their expected metric sets.
• The docs build passes with -W.

[coderabbitai]

Important

Draft PR not reviewed

Draft PRs are not automatically reviewed by default.

• Trigger a manual review

To automatically review draft PRs, update your CodeRabbit configuration:

reviews:
  auto_review:
    drafts: true

• Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

_{Comment @coderabbitai help to get the list of available commands.}