YAML Metadata Warning:empty or missing yaml metadata in repo card
Check out the documentation for more information.
pi05-real-jar-60-droid-refined-lora
LoRA fine-tune of pi0.5 (pi05_droid warm-start) on
IDEAS-Lab-Northwestern/real-jar-60-droid-refined — 60 real Franka teleop
trajectories of a jar-into-bowl transfer task (lid stays closed), DROID schema.
- Task prompt: "Move the jar into the bowl, ensuring that the lid remains securely closed during the transfer"
- Warm-start:
gs://openpi-assets/checkpoints/pi05_droid/params - LoRA:
gemma_2b_lora(rank 16) +gemma_300m_lora(rank 32), batch 4, EMA off. - Norm-stats: reuses
pi05_droid's bundled DROID assets (no recompute).
Checkpoint ladder
| Dir | Cumulative steps | Train loss |
|---|---|---|
10000/ |
10k | 0.0062 |
20000/ |
20k (saved at step 19999) | 0.0055 |
~298 steps/traj, ~4.4 epochs over 15k frames. Train loss converges ~0.005; eval-sweep both checkpoints.
Refined real setup
Trained on data from the refined real setup — refined wrist & main-camera
poses and gripper — distinct from earlier real datasets. Wrist = raw cam1,
exterior = raw cam0 (routed via --wrist-cam cam1). Collector: yypeng666.
Schema (DROID) / inference
exterior_image_1_left ← exterior cam0, wrist_image_left ← wrist cam1,
exterior_image_2_left zero-pad; joint_position(7)+gripper_position(1) state;
actions(8)=joint_velocity(7)+next gripper target(1). Serve with openpi
pi05_droid_finetune_lora. train_state/ omitted (inference-only). Eval on a
real Franka, not the SFT server.