---
license: apache-2.0
library_name: openpi
tags:
- robotics
- vision-language-action
- pi0
- aloha
- candy-garlic-pickup
pipeline_tag: robotics
---

# PI0 Candy and Garlic Pickup Model

This model is trained using the OpenPI framework for the task of picking up candy and garlic objects and placing them in a bowl.

## Model Details

- **Model Type**: PI0 (Physical Intelligence Zero)
- **Task**: Pick up candy and garlic, and put it in the bowl
- **Config**: `pi0_candy_garlic_pickup`
- **Training Steps**: 40,000
- **Action Horizon**: 50
- **Framework**: OpenPI

## Usage

```python
# Load the model using OpenPI
from openpi.training.config import get_config
from openpi.training import weight_loaders

# Get the config
config = get_config("pi0_candy_garlic_pickup")

# Load the model weights
weight_loader = weight_loaders.CheckpointWeightLoader("ry-5/pi0_candy_garlic")
# ... rest of your inference code
```

## Training Configuration

The model was trained with the following configuration:
- Repository: `ry-5/garlic_and_candy`
- Default prompt: "pick up the candy and garlic, and put it in the bowl"
- Save interval: 2000 steps
- Keep period: 20000 steps
- Action horizon: 50

## Files

- `params/`: Model parameters in Orbax format
- `train_state/`: Training state including optimizer state
- `assets/`: Associated assets including normalization statistics
- `_CHECKPOINT_METADATA`: Checkpoint metadata

## Citation

If you use this model, please cite the OpenPI framework.