Instructions to use OmakaseAI/pi05_yubi_ee_pickplace_v2 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- LeRobot
How to use OmakaseAI/pi05_yubi_ee_pickplace_v2 with LeRobot:
- Notebooks
- Google Colab
- Kaggle
pi05_yubi_ee_pickplace_v2
pi05 on the cleaned (v2) pick-and-place subtask dataset — the recommended
pickplace checkpoint (supersedes pi05_yubi_ee_pickplace).
- Dataset:
OmakaseAI/yubi_ee_pickplace_v2(708 train / 82 holdout, 54,830 frames). Built via the omakase-data-infra segment pipeline with the 2026-07-02 forensics fixes: stray-day sessions excluded, evidence-based segment selection (grasp cycle), tail raise-to-click gesture trimmed (~10 frames/ep), QC gate (only 2 eps dropped). Right-hand-only collection (left arm dims idle — recorded in metadata). - Actions: ABSOLUTE (relative=true + all-20-dim exclude). 20-dim bimanual EE, 3 cams, 7500 steps, full-param (vision frozen), 8x H100.
- wandb: omakase-robotics/galaxea_pi05_ee/runs/jsisyoyk
Serve via galaxea_lerobot serve_pi05_yubi_ee_bimanual.sh (MODEL_ID/DATASET_ID = this pair); client pi05_ee_inference. Open-loop MSE is not closed-loop success; on-robot eval required.
- Downloads last month
- 47
Model tree for OmakaseAI/pi05_yubi_ee_pickplace_v2
Base model
lerobot/pi05_base