> ## Documentation Index
> Fetch the complete documentation index at: https://mintlify.com/NVIDIA/Isaac-GR00T/llms.txt
> Use this file to discover all available pages before exploring further.

# SO-100 benchmark

> Fine-tuning and deployment for SO-100 robot arms

This guide shows how to fine-tune GR00T on datasets collected from the [SO-100](https://huggingface.co/docs/lerobot/en/so101) robot and deploy the model on real hardware.

## Dataset

### Data collection

To collect datasets via teleoperation, refer to the official LeRobot documentation:

[SO-100 teleoperation guide](https://huggingface.co/docs/lerobot/il_robots?teleoperate_so101=Command)

### Example dataset

**Dataset:** [izuluaga/finish\_sandwich](https://huggingface.co/datasets/izuluaga/finish_sandwich)

Visualize the dataset: [Dataset viewer](https://huggingface.co/spaces/lerobot/visualize_dataset?path=%2Fizuluaga%2Ffinish_sandwich%2Fepisode_0)

## Preparation

<Steps>
  <Step title="Convert dataset format">
    Convert from LeRobot v3 to v2 format:

    ```bash theme={null}
    uv run python scripts/lerobot_conversion/convert_v3_to_v2.py \
        --repo-id izuluaga/finish_sandwich \
        --root examples/SO100/finish_sandwich_lerobot
    ```
  </Step>

  <Step title="Copy modality configuration">
    ```bash theme={null}
    cp modality.json examples/SO100/finish_sandwich_lerobot/meta/modality.json
    ```
  </Step>
</Steps>

## Fine-tuning

Run the fine-tuning script using absolute joint positions:

```bash theme={null}
uv run bash examples/SO100/finetune_so100.sh
```

<Note>
  Feel free to experiment with relative joint positions by modifying the action modality configuration in `modality.json`.
</Note>

## Evaluation

### Open-loop evaluation

Evaluate the fine-tuned model against ground truth trajectories:

```bash theme={null}
uv run python gr00t/eval/open_loop_eval.py \
    --dataset-path examples/SO100/finish_sandwich_lerobot \
    --embodiment-tag NEW_EMBODIMENT \
    --model-path /tmp/so100_finetune/checkpoint-10000 \
    --traj-ids 0 \
    --action-horizon 16 \
    --steps 400
```

This generates visualizations comparing predicted actions against ground truth:

<img src="https://mintlify.s3.us-west-1.amazonaws.com/nvidia-isaac-gr00t/media/open_loop_eval_so100.png" alt="Open-loop evaluation results" />

### Closed-loop evaluation (real robot)

For deploying on real SO-100 hardware, see [eval\_so100.py](../../gr00t/eval/real_robot/SO100/eval_so100.py) for Policy API usage.

<Steps>
  <Step title="Setup client dependencies">
    ```bash theme={null}
    cd gr00t/eval/real_robot/SO100
    uv venv
    source .venv/bin/activate
    uv pip install -e . --verbose
    uv pip install --no-deps -e ../../../../
    ```
  </Step>

  <Step title="Start policy server">
    In Terminal 1:

    ```bash theme={null}
    uv run python gr00t/eval/run_gr00t_server.py \
        --model-path /tmp/so100_finetune/checkpoint-10000 \
        --embodiment-tag NEW_EMBODIMENT
    ```
  </Step>

  <Step title="Run evaluation client">
    In Terminal 2:

    ```bash theme={null}
    uv run python gr00t/eval/real_robot/SO100/eval_so100.py \
        --robot.type=so101_follower \
        --robot.port=/dev/ttyACM2 \
        --robot.id=orange_follower \
        --robot.cameras="{ wrist: {type: opencv, index_or_path: 2, width: 640, height: 480, fps: 30}, front: {type: opencv, index_or_path: 6, width: 640, height: 480, fps: 30}}" \
        --policy_host=localhost \
        --policy_port=5555 \
        --lang_instruction="cube into green bowl"
    ```
  </Step>
</Steps>

## Hardware configuration

When deploying on real hardware, ensure:

* Correct USB port for robot connection (`/dev/ttyACM*`)
* Camera indices match your hardware setup
* Camera resolution and FPS are compatible with your cameras
* Robot ID matches your configured follower arm

<Warning>
  Always test in a safe environment with emergency stop capabilities before running autonomous manipulation tasks.
</Warning>

## Additional resources

* [LeRobot SO-100 documentation](https://huggingface.co/docs/lerobot/en/so101)
* [Policy API guide](/getting-started/policy) - Understanding the policy interface
* [Fine-tune new embodiment](/getting-started/finetune-new-embodiment) - Custom robot configuration