PIv1

Training Config:

CONFIG = {
  # Dataset (pre-converted v3.0 format)
  "dataset_repo_id": "bdhillon/PI-0.5-11.19.2025-v3-quantiles",
  "dataset_root": os.path.expanduser("~/lerobot-training/dataset/PI-0.5-11.19.2025-v3-quantiles"),

  # Model
  "policy_type": "pi05",
  "pretrained_path": "lerobot/pi05_base",

  # HuggingFace upload settings
  "repo_id": "bdhillon/PIv1",
  "push_to_hub": True,

  # Training hyperparameters
  "batch_size": 4,
  "policy.dtype": "bfloat16",
  "policy.use_amp": True,
  "steps": 1500,           # ~3-4 epochs for 11 episodes with 6953 frames
  "eval_freq": 250,        # Evaluate every 250 steps
  "log_freq": 50,          # Log to WandB every 50 steps
  "save_freq": 250,        # Save checkpoint every 250 steps

  # Evaluation settings
  "eval_n_episodes": 5,
  "eval_batch_size": 5,    # Must be <= eval_n_episodes

  # Output
  "output_dir": "./PIv1",

  # Logging
  "wandb_enable": True,
}

---

π₀.₅ (Pi05) Policy

π₀.₅ is a Vision-Language-Action model with open-world generalization, from Physical Intelligence. The LeRobot implementation is adapted from their open source OpenPI repository.

For more details, see the Physical Intelligence π₀.₅ blog post.

---

Train From Scratch

lerobot-train \
  --dataset.repo_id=${HF_USER}/<dataset> \
  --policy.type=act \
  --output_dir=outputs/train/<desired_policy_repo_id> \
  --job_name=lerobot_training \
  --policy.device=cuda \
  --policy.repo_id=${HF_USER}/<desired_policy_repo_id>
  --wandb.enable=true

Writes checkpoints to outputs/train/<desired_policy_repo_id>/checkpoints/.

Evaluate / Run inference

lerobot-record \
  --robot.type=so100_follower \
  --dataset.repo_id=<hf_user>/eval_<dataset> \
  --policy.path=<hf_user>/<desired_policy_repo_id> \
  --episodes=10

Prefix the dataset repo with eval_ and supply --policy.path pointing to a local or hub checkpoint.

---