YAML Metadata Warning:empty or missing yaml metadata in repo card
Check out the documentation for more information.
xlvla-pi05 β RSS2027 pi0.5 baselines (XL-VLA latent and joint space)
pi0.5 (openpi, JAX) policies trained on DexVerse teleoperation demos (goal marker visible in rendering and evaluation).
One folder per experiment; each holds checkpoint/<step>/ (params + assets = norm stats, no train_state),
dataset/ (the LeRobot training set, tar) and config.yaml. LAD-latent models are in Dexverse-CSI/lad-pi05.
multitask/ 6 hands x 4 one-arm tasks (GB GP GK FH), one model per method, 20k steps
pi05_XL/ XL-VLA 6-hand VAE latent (32) action space
pi05/ joint-space action (baseline)
bimanual_pi05_XL/ 6 hands x 4 two-arm tasks (OFF SUK BLT BLC), latent per arm (82 dims)
bimanual_pi05/ same, joint space (62 dims)
zeroshot/pi05_XL/{GB,GP,GK,FH}/ one task, trained on 5 hands (shadow held out), 5k steps, evaluated on all 6
diagonal/pi05_XL/ multitask minus 4 held-out cells (shadow-GB, inspire-GP, sharpa-GK, wuji-FH)
Task abbreviations: GB GraspBleach, GP GraspPan, GK GraspKettle, FH FunctionalHammerStrike,
OFF OpenFlatFolder, SUK SlideUtilityKnife, BLT BimanualLiftTray, BLC BimanualLiftCarton.
Code: github.com/HyeonseokE/cross_dex (cross_dex/baselines/xlvla/dexverse_xlvla).
Inference Providers NEW
This model isn't deployed by any Inference Provider. π Ask for provider support