Skip to content

Install HASimulator and Get Public Data

This page is for researchers and developers who are setting up HA-VLN for the first time.

Your goal in this step is simple:

  • install the HA-VLN runtime and HASimulator
  • obtain the public data needed for development
  • verify that the environment can run before you start writing your own agent

What You Need

To develop an HA-VLN agent, you need:

  • the HA-VLN repository
  • the simulator stack used by HA-VLN
  • the public dataset under Data/
  • an available Python environment

For installation details, use these reference pages:

These pages use the Python 3.8 based setup path.

1. Clone the repository

git clone https://github.com/F1y1113/HA-VLN.git
cd HA-VLN

2. Prepare the Python environment

Follow Dependencies and Installation Steps to install the required system and Python packages.

3. Install the simulator stack

The installation flow includes the HA-VLN runtime together with Habitat-Sim and Habitat-Lab. Follow:

4. Prepare the public data under Data/

Place the public HA-VLN data under the repository Data/ directory. The full download sources, Matterport3D access note, extraction commands, and final layout are described in:

At minimum, your local setup should include the required Matterport3D Habitat assets, the public HA-R2R data, and the HAPS 2.0 human assets.

5. Verify the environment before agent development

Before writing or adapting your own agent, confirm both of the following:

  • the Python and simulator environment imports correctly
  • the HA-VLN runtime and data paths are available from the repository checkout

Basic environment checks

These checks only confirm that the base Python environment is usable:

python -c "import torch; print(torch.__version__)"
python -c "import habitat_sim; print('habitat-sim OK')"

HA-VLN runtime checks

Use a slightly stronger check from the repository root to confirm that the HA-VLN-specific packages and config path can be resolved through the local import-path setup used by the evaluation workflow:

python -c "import os, sys; repo = os.getcwd(); sys.path.insert(0, repo); sys.path.insert(0, os.path.join(repo, 'agent')); sys.path.insert(0, os.path.join(repo, 'agent', 'VLN-CE')); import HASimulator, habitat_extensions, vlnce_baselines; print('HA-VLN runtime OK')"

These checks are still lightweight. They do not prove that a full evaluation run will succeed, but they are much closer to the actual HA-VLN workflow than import-only checks for torch or habitat_sim.

Data-path sanity checks

Before moving on, make sure the required public paths are actually present under Data/.

For example, confirm that your checkout contains the expected public assets described in Data Download, including:

  • the HA-R2R dataset files
  • the HAPS 2.0 human assets
  • the required Matterport3D Habitat scene assets

What You Should Have at the End of This Step

When this step is complete, you should have:

  • a working HA-VLN repository checkout
  • a runnable simulator environment
  • the public dataset prepared under Data/
  • enough runtime support to begin developing your own agent

Next Step

Once installation and data preparation are finished, continue to: