Flower Calvin Abc

🚀 FlowerVLA - Vision-Language-Action Flow Model for CALVIN ABC

A pretrained FlowerVLA model for robotic manipulation, trained on the CALVIN ABC dataset. Flower is an efficient Vision-Language-Action Flow policy for robot learning with only 1B parameters.

📚 Documentation

Model Description

FlowerVLA is a novel architecture that:

• Utilizes half of Florence-2 for multi-modal vision-language encoding.
• Employs a novel transformer-based flow matching architecture.
• Offers an efficient, versatile VLA policy with only approximately 1B parameters.

Model Performance

This checkpoint contains weights for the CALVIN ABC challenge and currently ranks 1 with the following results:

Input/Output Specifications

Inputs

• RGB Static Camera: (B, T, 3, H, W) tensor
• RGB Gripper Camera: (B, T, 3, H, W) tensor
• Language Instructions: Text strings

Outputs

• Action Space: (B, T, 7) tensor representing delta EEF actions

💻 Usage Examples

Basic Usage

obs = {
    "rgb_obs": {
        "rgb_static": static_image,
        "rgb_gripper": gripper_image
    }
}
goal = {"lang_text": "pick up the blue cube"}
action = model.step(obs, goal)

Advanced Usage

Check out our full model implementation on Github todo and follow the instructions in the readme to test the model on one of the environments.

🔧 Technical Details

Configuration

• Optimizer: AdamW
• Learning Rate: 2e-5
• Weight Decay: 0.05

📄 License

This model is released under the MIT license.

Citation

@inproceedings{ reuss2025flower, # Add citation when available }