| --- |
| datasets: |
| - hellorobotinc/kitchen_cabinet_diagonal |
| pipeline_tag: robotics |
| tags: |
| - LeRobot |
| --- |
| |
| ## Model Card for Diffusion Policy / Kitchen cabinet opening - Diagonal configuration |
| Diffusion Policy trained for [Stretch 3](https://hello-robot.com/stretch-3-product) using [LeRobot](https://huggingface.co/lerobot). Trained to open kitchen cabinet from diagonal orientation in Yi-Che (Jensen) Huang's home during his Summer 2024 internship. |
| ## Training Details |
| Model trained on Dex-Teleop demonstrations collected with [stretch_ai](https://github.com/hello-robot/stretch_ai) |
|
|
| - Dataset: [hellorobotinc/kitchen_cabinet_diagonal](https://huggingface.co/datasets/hellorobotinc/kitchen_cabinet_diagonal) |
| - Number of episodes: 50 |
| - Number of epochs: 1000 |
| - Checkpoint: latest checkpoint |
| - Training time: ~11 hours on RTX 3090 |
|
|
| ## Input/Output |
| Inputs: |
|
|
| - Head RGB (320x240) |
| - Gripper RGB (240x320) |
| - 9 dim joint state observations (x, y, theta, lift, arm, roll, pitch, yaw, gripper) |
|
|
| Output: |
| - 9 dim joint state actions (x, y, theta, lift, arm, roll, pitch, yaw, gripper) |
| - **optional extra dim for task progress (included in dataset, but not trained into this model) |
| |
| Specific training configs can be found in config.json and config.yaml |
