alexhegit/so101-simstudio-lab01-pnp-pi05-agent193

SO-101 Lab01 Pick-and-Place — pi05 (193 episodes, 10K)

pi05 fine-tune from ant0nh/pi05_pnp_425_25k (itself a fine-tune of lerobot/pi05_base on a real SO-follower pick-and-place set). train_expert_only=true, use_relative_actions=false, dtype bfloat16. Cameras: camera_wrist → observation.images.wrist, camera_top → observation.images.front. State and action are 6-D absolute joint .pos.

Training data: alexhegit/so101-simstudio-lab01-pnp-agent-yaw-v4-actonly-follow2

Lab01-PnP-Home20: 4/20, 6/20, 2/20 = 12/60. Reach 60/60, close 39/60, lift 15/60. The hole is lift after close. The later full-unfreeze run is a different repo, alexhegit/so101-simstudio-lab01-pnp-pi05-agent193-unfreeze, at 34/60. Comparison: home20.md.

Checkpoint 010000. Chunk / n_action_steps 50. Sync inference.

from lerobot.policies.pi05.modeling_pi05 import PI05Policy

policy = PI05Policy.from_pretrained(
    "alexhegit/so101-simstudio-lab01-pnp-pi05-agent193"
)