> ## 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.

# PointNav benchmark

> Point navigation tasks with COMPASS-generated datasets

This guide shows how to fine-tune GR00T for point navigation tasks using datasets generated by [COMPASS](https://github.com/NVlabs/COMPASS), NVIDIA's framework for learning navigation policies.

## Dataset preparation

To generate and prepare the dataset, follow the [COMPASS GR00T post-training guide](https://github.com/NVlabs/COMPASS?tab=readme-ov-file#gr00t-post-training-with-compass-datasets):

<Steps>
  <Step title="Train residual RL specialists">
    Use COMPASS to train task-specific RL policies.
  </Step>

  <Step title="Collect distillation data">
    Collect specialist distillation data from trained policies.
  </Step>

  <Step title="Convert to LeRobot format">
    Convert HDF5 dataset to GR00T LeRobot format:

    ```bash theme={null}
    python scripts/hdf5_to_lerobot_episodic.py
    ```
  </Step>
</Steps>

### Quick start dataset

For a quick start, a pre-collected G1 robot dataset is available:

```bash theme={null}
wget https://huggingface.co/nvidia/COMPASS/blob/main/gr00t_post_training_g1.zip
unzip gr00t_post_training_g1.zip
```

## Modality configuration

The point navigation task uses the following modalities defined in `modality.json`:

### Input modalities

| Modality | Key                | Indices   | Dimension | Description                                            |
| -------- | ------------------ | --------- | --------- | ------------------------------------------------------ |
| Video    | `ego_view`         | -         | H×W×3     | Ego-centric RGB camera image                           |
| State    | `speed`            | \[0, 1)   | 1         | Robot forward speed                                    |
| State    | `route`            | \[1, 41)  | 40        | Route segments in robot frame (10 segments × 4 values) |
| State    | `goal_heading`     | \[41, 43) | 2         | Goal heading direction (cos θ, sin θ)                  |
| Language | `task_description` | -         | -         | "Robot Navigation Task"                                |

### Output modalities

| Modality | Key       | Indices | Dimension | Description                   |
| -------- | --------- | ------- | --------- | ----------------------------- |
| Action   | `vel_cmd` | \[0, 3) | 3         | Velocity command (vx, vy, ωz) |

<Note>
  The route modality encodes 10 waypoint segments, with each segment represented by 4 values: x\_start, y\_start, x\_end, y\_end in the robot's local frame.
</Note>

## Fine-tuning

Run the fine-tuning script after updating paths:

```bash theme={null}
uv run bash examples/PointNav/finetune_point_nav.sh
```

Update the script with your configuration:

* `--dataset-path`: Path to the converted LeRobot format dataset
* `--output-dir`: Directory to save checkpoints

## Evaluation

<Steps>
  <Step title="Launch inference server">
    Start the GR00T policy server:

    ```bash theme={null}
    uv run python gr00t/eval/run_gr00t_server.py \
        --model-path <path/to/checkpoint> \
        --embodiment-tag NEW_EMBODIMENT \
        --device cuda:0 \
        --host 0.0.0.0 \
        --port 8888
    ```
  </Step>

  <Step title="Run COMPASS evaluation">
    Follow the [COMPASS evaluation instructions](https://github.com/NVlabs/COMPASS?tab=readme-ov-file#gr00t-post-training-with-compass-datasets) to evaluate the fine-tuned model.
  </Step>
</Steps>

## Results

Task success rate on 640 randomized test cases:

| Model              | In-distribution | Out-of-distribution |
| :----------------- | :-------------: | :-----------------: |
| GR00T N1.6         |    **86.3%**    |      **76.5%**      |
| GR00T N1.5         |      86.1%      |        77.6%        |
| COMPASS (baseline) |      84.7%      |        45.6%        |

<Note>
  GR00T significantly outperforms the COMPASS baseline on out-of-distribution scenarios (76.5% vs 45.6%), demonstrating strong generalization capabilities.
</Note>

## Additional resources

* [COMPASS repository](https://github.com/NVlabs/COMPASS) - Dataset generation and evaluation
* [COMPASS documentation](https://github.com/NVlabs/COMPASS?tab=readme-ov-file#gr00t-post-training-with-compass-datasets) - Detailed integration guide