xNav is an ambitious project aiming to realize a true dual-system Vision-Language-Action (VLA) model for intelligent language navigation.
We use uv to manage this project. Use following command to setup the environment.
uv sync
uv pip install -e .Run following command to export model with GR00T action head and DepthAnythingV2 ViT (change the path in the scripts to real path)
uv run xnav/module/export_xnav_dit.pySystem 1
Use following command to train system1
uv run accelerate launch scripts/train_xnav_dit.py --model-name-or-path path/to/XNavDit --dataset-path path/to/dataset