polaris_pi05_sft

Fine-tuned π0.5 (pi05) DROID policy for the Polaris real2sim stack.

• Base config: pi05_droid (openpi)
• Recipe: pi05_droid_jointpos_lerobot_finetune_polaris (joint-position SFT on the Polaris LeRobot dataset)
• Checkpoint step: 19999 (final)
• Contents: inference-ready only — params/ + assets/droid/norm_stats.json. The optimizer train_state/ is not included, so this checkpoint serves but cannot resume training.

Usage

Download the repo and point the Polaris openpi client at it:

huggingface-cli download Makolon0321/polaris_pi05_sft --local-dir /path/to/ckpt
export POLARIS_PI05_CKPT=/path/to/ckpt

The repo root acts as the openpi checkpoint dir; create_trained_policy reads params/ and assets/ from it:

from openpi.policies import policy_config
from openpi.training import config as _config

train_cfg = _config.get_config("pi05_droid")
policy = policy_config.create_trained_policy(train_cfg, "/path/to/ckpt")