arkojit1/pi05_pick_block_eef_delta

π₀.₅ fine-tuned on Ameyapores/pick_block_eef_delta

lerobot/pi05_base fine-tuned with LeRobot (--policy.type=pi05) on Ameyapores/pick_block_eef_delta — 35 Franka episodes, 1 task, two 224×224 cameras (cam0, cam2; the third π₀.₅ image slot is padded, empty_cameras=1), 4-D state, 4-D end-effector-delta action.

This is the step 1,500 checkpoint (~47 epochs) of the baseline run, selected as the lowest held-out eval loss.

Trainable	action expert only (train_expert_only=true; SigLIP + Gemma-2B frozen)
Global batch	256 (32/GPU × 8 MI300X)
LR	2.5e-5 peak, cosine to 2.5e-6, bf16
Normalisation	quantile (state and action)
Augmentation	LeRobot image transforms on train frames
chunk_size / n_action_steps	50 / 50
Eval split	last 4 of 35 episodes held out
Eval loss	0.1031 (flow-matching loss on the held-out episodes)

Eval loss is a training-objective number on 4 held-out episodes, not a task success rate; with a holdout this small, nearby checkpoints (steps ~1,000–2,000) are statistically indistinguishable from this one.

from lerobot.policies.pi05.modeling_pi05 import PI05Policy
policy = PI05Policy.from_pretrained("arkojit1/pi05_pick_block_eef_delta")