libero_seen_pretrain
Collection
Seen-suite (libero_90) pretraining for LIBERO-Goal few-shot: SmolVLA action BC + action-free latent dynamics (IDM/FDM). • 5 items • Updated
SmolVLA · libero_90 · 100k
Action-labeled behavior-cloning pretrain of lerobot/smolvla_base
on nvidia/libero_90 (LeRobot v3), 100k steps.
Use as --policy.path for few-shot LIBERO-Goal fine-tunes in this project.
# example
lerobot-train --policy.path=<this-repo> ...
Vision encoder frozen; action expert + state_proj trained. No LIBERO-Goal demos in this checkpoint.
- Downloads last month
- 18