rllab-postech/pretrain_aiworker_bh5_sh5_hand_lance
rllab-postech/pretrain_aiworker_bh5_sh5_hand_lance Merged BH5/SH5 dual-hand 54D pretraining dataset in RLLAB published Lance layout. Tables Table Purpose data/episodes.lance Published episode table, one row per episode, no video blob columns. data/train_episodes.lance Training trajectory table named by manifest.json.primary_training_table; no video blob columns. data/frames.lance Frame-level QA/index table with remapped global frame indices.… See the full description on the dataset page: https://huggingface.co/datasets/rllab-postech/pretrain_aiworker_bh5_sh5_hand_lance.
rllab-postech/pretrainaiworkerbh5sh5hand_lance
Merged BH5/SH5 dual-hand 54D pretraining dataset in RLLAB published Lance layout.
Tables
Summary
- Source datasets: 9
- Episodes: 189
- Frames: 73348
- Videos: 189
- Action/state dim: 54 / 54
- FPS values: 10.0
- Format: rllabpublishedlancedatasetv2 / schema 2.0
- Blob storage: lance.blob.v2 (inlinebytesonly)
Quality Filters
- Min FPS: 10.0
- Max FPS: 30.0
- Min episode frames: 30
- Max episode frames: disabled
- Drop wrist-only episodes: False
- Required camera keys: none
- Require any camera keys: none
- Kept camera keys: all
- Filtered episodes: 0
Each Lance row carries provenance columns such as source_dataset, source_repo_id, source_dataset_url, source_local_path, source_episode_index, source_robot_type, and pretrain_tier. data/videos.lance also keeps source_video_table, source_media_id, and source_relative_path while storing the canonical copied MP4 blob used by viewers and training.
Sources
Regenerate
PYTHONPATH=. ./.venv/bin/python scripts/build_pretrain_19d_lance.py --overwrite