Installation Steps

Official HA-VLN repository: https://github.com/F1y1113/HA-VLN

The following installation flow is the default setup for HA-VLN.

This documentation uses a Python 3.8 based environment as the default path for better compatibility on modern GPUs and modern Linux environments.

1. Clone Repository

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

2. Create the Runtime Environment

conda create -n havlnce python=3.8 gcc_linux-64=11 gxx_linux-64=11 sysroot_linux-64=2.17 -c conda-forge -y
conda activate havlnce

conda install -c "nvidia/label/cuda-11.8.0" cuda-toolkit -y
pip install torch==2.0.1+cu118 torchvision==0.15.2+cu118 --index-url https://download.pytorch.org/whl/cu118

export CC=$CONDA_PREFIX/bin/x86_64-conda-linux-gnu-gcc
export CXX=$CONDA_PREFIX/bin/x86_64-conda-linux-gnu-g++
export CUDA_HOME=$CONDA_PREFIX

3. Install Habitat-Sim and Habitat-Lab

conda install -c conda-forge python-lmdb libxcrypt libffi -y
conda install -c aihabitat -c conda-forge "habitat-sim=0.1.7=*headless*" -y

git clone --branch v0.1.7 https://github.com/facebookresearch/habitat-lab.git
cd habitat-lab
pip install msgpack "numpy<1.24.0" tensorboard
sed -i 's/tensorflow==1.13.1/# tensorflow==1.13.1/g' habitat_baselines/rl/requirements.txt
pip install -r requirements.txt
pip install -r habitat_baselines/rl/requirements.txt
python setup.py develop --all
cd ..

4. Install GroundingDINO

cd HASimulator
git clone https://github.com/IDEA-Research/GroundingDINO.git
cd GroundingDINO

rm -rf build/ dist/ *.egg-info
pip install "safetensors==0.3.1" "huggingface-hub==0.16.4" \
    "tokenizers==0.13.3" "transformers==4.30.2" "timm==0.9.2"
sed -i 's/supervision.*/supervision==0.11.1/g' requirements.txt
pip install ninja
pip install -e .
cd ../..

If GroundingDINO compilation hits __int128 typedef errors, patch the conda sysroot header before retrying:

SYSROOT_TYPES="$CONDA_PREFIX/x86_64-conda-linux-gnu/sysroot/usr/include/linux/types.h"
chmod +w "$SYSROOT_TYPES"
sed -i 's/typedef __signed__ __int128/\/\/ typedef __signed__ __int128/g' "$SYSROOT_TYPES"
sed -i 's/typedef unsigned __int128/\/\/ typedef unsigned __int128/g' "$SYSROOT_TYPES"

5. Install Agent-Side Python Requirements

pip install -r agent/VLN-CE/requirements.txt
pip install "Pillow<9.0.0" setuptools webdataset==0.1.40

# Reassert the expected PyTorch stack in case pip changed it
pip install torch==2.0.1+cu118 torchvision==0.15.2+cu118 --index-url https://download.pytorch.org/whl/cu118

6. Apply the Gym Compatibility Patch

Modern versions of gym do not allow Discrete(0) initialization in the old code path used here.

sed -i 's/spaces.Discrete(0)/spaces.Discrete(4)/g' habitat-lab/habitat/tasks/vln/vln.py

7. Verify Runtime

At minimum, verify these checks:

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

FAQ: Why does this page use Python 3.8 instead of the original README's Python 3.7?

The original README reflects an older environment. These docs use Python 3.8 as the default path because it is a more practical compatibility baseline for newer GPUs and newer CUDA/PyTorch combinations.

The important takeaway is simple: follow the Python 3.8 path in this documentation unless you have a specific reason to maintain an older legacy environment.