Repository navigation
feat: arrange-3-flowers task (DOF layout fix, data conversion, training/deploy docs)Feat/arrange 3 flowers - #118
Merged
StarrickLiu merged 7 commits intoAug 31, 2026
Conversation
Match LAYOUT_SKIP_KEYS across loader, norm stats, and inference so datasets with extra velocity/height/head DOFs can train on arm-only layouts. Co-authored-by: Cursor <cursoragent@cursor.com>
Add LeRobot-based arrange_3_flowers training doc, example configs, and rtx5090 install/run scripts with env-based path layout for inference. Co-authored-by: Cursor <cursoragent@cursor.com>
Link arrange_3_flowers training example and RTX 5090 deploy docs for newcomers. Co-authored-by: Cursor <cursoragent@cursor.com>
Align ground-truth action dimensions with model predictions when datasets store extra DOF padding beyond the effective arm-only dof_config. Co-authored-by: Cursor <cursoragent@cursor.com>
Cherry-pick DOF layout fix, training/deploy docs, and open-loop eval fix from fix/lerobot-dof-layout onto upstream/main. Add x2robot2lerobot conversion script with template configs, sanitize internal paths to placeholders, and stop tracking machine-specific rtx5090/local/env.sh. Co-authored-by: Cursor <cursoragent@cursor.com>
Drop the hardcoded Alibaba OSS URL for Mellanox RDMA user-space libs so the open-source example no longer depends on an internal artifact host. Co-authored-by: Cursor <cursoragent@cursor.com>
Fix truncated Qwen path placeholders, use generic wallx conda default, and point NOTES.md at env.example instead of the gitignored env.sh. Co-authored-by: Cursor <cursoragent@cursor.com>
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.
Summary
action_padding(loader, norm stats, inference).x2robot2lerobot.pyand template configs for x2robot → LeRobot v3 conversion.action_paddingfrom ground-truth actions.Motivation
When the dataset stores extra DOFs (velocity/height/head) but training only predicts dual-arm actions with
action_padding, the loader, norm stats, and serving must agree on the effective layout.Test plan
python scripts/compute_norm_stats.py --train_config workspace/example/arrange_3_flowers_wrc_red.yml(with local paths)arrange_3_flowers_wrc_red_smoke.ymlbash workspace/rtx5090/run_server.shand check/healthzNotes
<path/to/...>placeholders per CONTRIBUTING.md.workspace/rtx5090/local/env.shis gitignored.