# Pi0-real-robot

PyTorch `pi0` policy fine-tuned on the local UR5 real-robot LeRobot-format dataset:

- dataset: `ur5_lab_test_tube_camera_shifts`
- training config name: `pi0_ur5_real_robot_pytorch_baseline`
- base camera view: `observation.images.context_left_rgb`
- wrist camera view: `observation.images.wrist_right_rgb`
- base model init: `/scratch/yz11445/pi0_base`
- included checkpoint steps: `20000`, `25000`, `30000`

## Included Files

- `config.json`: base Pi0 model config copied from the initialization checkpoint
- `model_architecture_config.json`: fine-tuned architecture settings used by this run
- `training_config_summary.json`: training/data/run summary for this release
- `assets/ur5_lab_test_tube_camera_shifts/norm_stats.json`: normalization statistics
- `checkpoints/<step>/`: checkpoint snapshots with `model.safetensors`, `metadata.pt`, and copied assets

## Inference

Serve any checkpoint with:

```bash
uv run scripts/serve_policy.py policy:checkpoint \
  --policy.config=pi0_ur5_real_robot_pytorch_baseline \
  --policy.dir=/path/to/Pi0-real-robot/checkpoints/30000
```

Replace `30000` with one of `20000`, `25000`, or `30000`.

## Notes

- The policy loader uses the code-defined training config `pi0_ur5_real_robot_pytorch_baseline` from `src/openpi/training/config.py`.
- Normalization stats are loaded from `assets/ur5_lab_test_tube_camera_shifts/norm_stats.json` inside each checkpoint directory.
- Tokenizer assets are not bundled in this release directory. In this codebase, the Pi0 tokenizer is loaded at runtime from external sources referenced in `src/openpi/models/tokenizer.py`.