fix: support lerobot >= 0.6.0 in the hw_to_dataset_features import seam#21
Merged
Conversation
lerobot 0.6.0 moved hw_to_dataset_features from
lerobot.datasets.feature_utils to lerobot.utils.feature_utils, breaking
policy loading against 0.6.x. Extend the version seam to try the new
location first, falling back to the 0.5.1 and 0.5.0 layouts.
Verified live on hardware against lerobot 0.6.1 (SO-101 follower +
smolvla_base end-to-end). Note 0.6.x also gates previously-bundled deps
behind extras at import/load time (grpcio via lerobot[async],
datasets via lerobot[dataset]) — install those alongside.
Test fakes now model all three layouts via feature_layout
("v060" default / "v051" / "v050"), with explicit fallback tests for
both older layouts.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
aris-zhu
force-pushed
the
lerobot-06-seam
branch
from
July 14, 2026 16:08
8a06aa1 to
739aca4
Compare
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What
lerobot 0.6.0 moved
hw_to_dataset_featuresfromlerobot.datasets.feature_utilstolerobot.utils.feature_utils, which breaksLeRobotPolicyloading against 0.6.x (both prior import paths raise). This extends_import_hw_to_dataset_features()to try the 0.6 location first, then fall back to the 0.5.1 and 0.5.0 layouts. The call-site signature is unchanged across versions ((hw_features, prefix, use_video)).Why now
Hit in the field: running this plugin against a lerobot
maincheckout (0.6.1) on an SO-101 follower. With this fix the full stack (SO-101 +smolvla_base, in-process inference) runs end-to-end on hardware against 0.6.1.Tests
_install_lerobot_fakesnow models the three layouts viafeature_layout("v060"default /"v051"/"v050"); the default flip means most tests exercise the new path, with explicit fallback tests for both older layouts. Gates: ruff, ruff format, pytest (82 passed, 100% coverage), mypy strict — all green locally.Notes / out of scope
lerobotextra keeps its<0.6cap: 0.6.x also gates previously-bundled deps behind extras that surface lazily at policy-load time (grpciovialerobot[async]at import oflerobot.async_inference,datasetsvialerobot[dataset]). Users on 0.6.x must install those alongside; lifting the cap properly (extra-of-extras + seam CI matrix against both majors) is a follow-up.🤖 Generated with Claude Code