pi05-droid-franka-lora-10k

pi05-droid-franka-lora-10k

Intermediate checkpoint (step 10,000 / 20,000) of the same fine-tune as pi05-droid-franka-lora (the final step-19,999 checkpoint lives there). Kept because the training config used keep_period=5,000, so step 10,000 was retained for evaluation.

Fine-tuned pi0.5 (openpi) checkpoint for the TASL Franka FR3 bench. LoRA fine-tuned on 16 GELLO-teleoperated episodes (4,262 frames, 15 Hz) of "pick blue cup and put into red cup" pick-and-place on a Franka Research 3 with Robotiq 2F-85 gripper and dual ZED cameras. Base: pi05_droid (Physical Intelligence, DROID-pretrained). Norm-stats are the original DROID stats (asset_id=droid) and must be reused at inference.

Layout

• params/ — orbax (JAX) model weights
• assets/ — norm stats (droid/norm_stats.json)
• train_state/ — optimizer state (not needed for inference)
• _CHECKPOINT_METADATA — orbax checkpoint metadata

Serve with openpi

python scripts/serve_policy.py --port 8000 \
  policy:checkpoint \
  --policy.config=pi05_droid_franka_lora \
  --policy.dir=<path to this repo after download>

Same pi05_droid_franka_lora config as the step-19,999 checkpoint (same training run, earlier step).