junn0103/turbovla-so101-pen-basket

TurboVLA SO-101 — "pick the pen and put it in basket"

TurboVLA policy (lerobot_policy_turbovla_so101) trained on a real SO-101 arm dataset (40 episodes / 23,960 frames @ 30fps, single wrist camera, 6-D absolute joint-position actions).

• —Policy: turbovla_so101 (0.2B total params, ~14M trainable)
• —Vision backbone: facebook/dinov2-base (frozen; DINOv3 substitute — ungated)
• —Language backbone: google-bert/bert-base-uncased (frozen)
• —Task string: "pick the pen and put it in basket"
• —Training: 80k steps, batch 64, LR 5e-5 (10k warmup, cosine decay), L1 loss, 1× H200, 2h11m
• —Final train loss: 0.022 (checkpoint at step 80,000)
• —Curves: logged every 200 steps; full run on wandb

Usage

pip install lerobot_policy_turbovla_so101   # Python >= 3.12

lerobot-record \
  --robot.type=so101_follower ... \
  --policy.path=<this repo id>

Trained with lerobot-train --policy.type=turbovla_so101 on lerobot 0.6.1.