YAML Metadata Warning:empty or missing yaml metadata in repo card
Check out the documentation for more information.
MetaWorld MT50 LeRobot post-training data
Repository ID: Mark-ZJTang/metaworld_mt50.
LeRobot v2-format downstream fine-tuning dataset with 2,500 episodes, 204,806 frames, and 49 task IDs. This is a training dataset; MetaWorld MT50 simulator evaluation itself does not read it.
- Split:
train(0:2500), 80 FPS, three chunks. observation.image: image,[3, 480, 480].observation.state: float32,[4].observation.environment_state: float32,[39].action: float32,[4](x,y,z,gripper).- Per-frame reward/success, task, episode, frame, and global indices are retained.
Restore it to the ALAM repository's loader-compatible relative path and verify both downstream datasets with:
.venvs/publish/bin/python workflows/publishing/download_huggingface.py \
--artifact metaworld_mt50_lerobot
.venvs/pi0/bin/python workflows/pi0_post_training/tests/test_datasets.py
The resulting root is data/lerobot/metaworld_mt50; the public fine-tuning
wrapper supplies its parent as HF_LEROBOT_HOME.
Before publication, verify the provenance and redistribution rights of generated trajectories and bundled task assets, then add a license and generation details.
- Downloads last month
- 434