Fix aloha real-world datasets

[Cadene]

Review filters

What this does

Core:

• Some real world datasets had issues because of gdown, so I uploaded all raw dataset to the hub, to then being able to download them from the hub and getting rid of gdown.
• As a result 2 issues have been addressed:
  • aloha_static_thread_velcro videos are actually aloha_static_tape videos #168
  • --repo-id for mobile aloha datasets #187

Addition:

• Add safetensors to git lfs
• For every sim dataset, add an image dataset to be equivalent the original (instead of video which is compressed). Example: "lerobot/pusht" -> "lerobot/pusht_image"
• Set CODEBASE_VERSION v1.4 (for backward compatibility)
• Add hf-transfer as extra to download/upload faster with export HF_HUB_ENABLE_HF_TRANSFER=1
• Fix an episode_index issue affecting all the aloha datasets. As a result:
  • the order of episodes changed in all aloha datasets (sim included)
  • so all aloha sim datasets have been reuploaded
  • but we had to reupload dataset tests artifacts as well since episode_index=0 is not the same anymore
  • but we had to reupload policy tests artifacts as well since they were using episode_index=0
• Add *.mp4 *.json *.safetensors *.arrow as git-lfs. Enables git-lfs in CI.
• Add velocity and effort to all aloha (real-world) datasets @haixuanTao

How it was tested

• Manually checked all episodes are inside raw dataset stored on the hub in: https://huggingface.co/cadene

• Downloaded all datasets and visualized the first episode of each (I had to code a faster visualization than rerun: Add visualize_dataset_html with http.server #188)

• Double-checked backward compatibility with previous datasets

• Visualized random episodes of all datasets
  https://github.com/huggingface/lerobot/assets/4681518/07f71279-2d8c-409d-8f71-a3f8c1c29fb5

• Checked that the order of episodes had changed as expected for aloha sim datasets.

  • Frame index 0 from the old dataset
    
  • Frame index 0 from the new dataset (as expected, the episode is not the same)
    

How to checkout & try?

• Upload new/updated datasets (done on dgx):

datasets=(
  "pusht_image"
)
for dataset in "${datasets[@]}"; do
    python lerobot/scripts/push_dataset_to_hub.py \
    --data-dir /raid/remi_cadene/data \
    --dataset-id $dataset \
    --raw-format pusht_zarr \
    --community-id lerobot \
    --dry-run 1 \
    --save-to-disk 0 \
    --save-tests-to-disk 1 \
    --video 0;rm -rf /raid/remi_cadene/data/${dataset}_raw
done

datasets=(
  "xarm_lift_medium_image"
  "xarm_lift_medium_replay_image"
  "xarm_push_medium_image"
  "xarm_push_medium_replay_image"
)
for dataset in "${datasets[@]}"; do
    python lerobot/scripts/push_dataset_to_hub.py \
    --data-dir /raid/remi_cadene/data \
    --dataset-id $dataset \
    --raw-format xarm_pkl \
    --community-id lerobot \
    --dry-run 1 \
    --save-to-disk 0 \
    --save-tests-to-disk 1 \
    --video 0;rm -rf /raid/remi_cadene/data/${dataset}_raw
done

datasets=(
  "aloha_sim_insertion_human_image"
  "aloha_sim_insertion_scripted_image"
  "aloha_sim_transfer_cube_human_image"
  "aloha_sim_transfer_cube_scripted_image"
)
for dataset in "${datasets[@]}"; do
    python lerobot/scripts/push_dataset_to_hub.py \
    --data-dir /raid/remi_cadene/data \
    --dataset-id $dataset \
    --raw-format aloha_hdf5 \
    --community-id lerobot \
    --dry-run 0 \
    --save-to-disk 0 \
    --save-tests-to-disk 1 \
    --video 0;rm -rf /raid/remi_cadene/data/${dataset}_raw
done

datasets=(
  "aloha_sim_insertion_human"
  "aloha_sim_insertion_scripted"
  "aloha_sim_transfer_cube_human"
  "aloha_sim_transfer_cube_scripted"
  "aloha_mobile_cabinet"
  "aloha_mobile_chair"
  "aloha_mobile_elevator"
  "aloha_mobile_shrimp"
  "aloha_mobile_wash_pan"
  "aloha_mobile_wipe_wine"
  "aloha_static_battery"
  "aloha_static_candy"
  "aloha_static_coffee"
  "aloha_static_coffee_new"
  "aloha_static_cups_open"
  "aloha_static_fork_pick_up"
  "aloha_static_pingpong_test"
  "aloha_static_pro_pencil"
  "aloha_static_screw_driver"
  "aloha_static_tape"
  "aloha_static_thread_velcro"
  "aloha_static_towel"
  "aloha_static_vinh_cup"
  "aloha_static_vinh_cup_left"
  "aloha_static_ziploc_slide"
)
for dataset in "${datasets[@]}"; do
    export HF_HUB_ENABLE_HF_TRANSFER=1
    export HF_DATASETS_CACHE=/raid/remi_cadene/.cache/huggingface/datasets
    python lerobot/scripts/push_dataset_to_hub.py \
    --data-dir /raid/remi_cadene/data \
    --dataset-id $dataset \
    --raw-format aloha_hdf5 \
    --community-id lerobot \
    --dry-run 0 \
    --save-to-disk 0 \
    --save-tests-to-disk 1 \
    --video 1;rm -rf /raid/remi_cadene/data/${dataset}_raw
done

• Just checkout-ed v1.4 branch on datasets that didnt need update:

from huggingface_hub import create_branch
dataset_ids = ["pusht", "xarm_lift_medium", "xarm_lift_medium_replay", "xarm_push_medium", "xarm_push_medium_replay"]
for dataset_id in dataset_ids:
    create_branch(f"lerobot/{dataset_id}", repo_type="dataset", branch="v1.4")

• Visualize dataset with this PR (which has been rebased on top of user/rcadene/2024_05_11_fix_aloha_static_mobile_datasets): Add visualize_dataset_html with http.server #188

export HF_HUB_ENABLE_HF_TRANSFER=1
datasets=(
  "pusht_image"
  "xarm_lift_medium_image"
  "xarm_lift_medium_replay_image"
  "xarm_push_medium_image"
  "xarm_push_medium_replay_image"
  "aloha_sim_insertion_human_image"
  "aloha_sim_insertion_scripted_image"
  "aloha_sim_transfer_cube_human_image"
  "aloha_sim_transfer_cube_scripted_image"
  "pusht"
  "xarm_lift_medium"
  "xarm_lift_medium_replay"
  "xarm_push_medium"
  "xarm_push_medium_replay"
  "aloha_sim_insertion_human"
  "aloha_sim_insertion_scripted"
  "aloha_sim_transfer_cube_human"
  "aloha_sim_transfer_cube_scripted"
  "aloha_mobile_cabinet"
  "aloha_mobile_chair"
  "aloha_mobile_elevator"
  "aloha_mobile_shrimp"
  "aloha_mobile_wash_pan"
  "aloha_mobile_wipe_wine"
  "aloha_static_battery"
  "aloha_static_candy"
  "aloha_static_coffee"
  "aloha_static_coffee_new"
  "aloha_static_cups_open"
  "aloha_static_fork_pick_up"
  "aloha_static_pingpong_test"
  "aloha_static_pro_pencil"
  "aloha_static_screw_driver"
  "aloha_static_tape"
  "aloha_static_thread_velcro"
  "aloha_static_towel"
  "aloha_static_vinh_cup"
  "aloha_static_vinh_cup_left"
  "aloha_static_ziploc_slide"
  "umi_cup_in_the_wild"
)
for dataset in "${datasets[@]}"; do
  python lerobot/scripts/visualize_dataset.py --repo-id lerobot/${dataset} --episode-indices 0 --serve 0
done;

[AdilZouitine]

LGTM 🤗

[alexander-soare]

Approved.