dreamtacvla
Checkpoints for the DreamTacVLA policy trained on bimanual Dobot X-Trainer data with vision + tactile observations (recorded 2025-11-07).
Datasets: michaelyeah7/dobot_xtrainer_tac_1107
Subfolders
| Folder | Task | Notes |
|---|---|---|
dreamtacvla_gearassemb_1107/ |
Gear assembly | full training run |
dreamtacvla_peginhole_1107/ |
Peg-in-hole | full training run |
dreamtacvla_usb_vitg_1107/ |
USB insertion | ViT-G encoder variant |
Each subfolder contains:
config.pkl— training configdataset_stats.pkl— normalization stats from the training datasetpolicy_step_*_seed_42.ckpt— checkpoints at various training stepspolicy_last.ckpt— last checkpoint (where available)
Loading
from huggingface_hub import snapshot_download
path = snapshot_download(
repo_id="michaelyeah7/dreamtacvla",
allow_patterns="dreamtacvla_peginhole_1107/*", # or any other subfolder
)