GN-BAE

GN-BAE is the companion navigation policy checkpoint for GN0, a unified research framework for Generation, Evaluation, and Policy Learning in Vision-and-Language Navigation (VLN).

GN0 builds on 3D Gaussian Splatting (3DGS) to connect realistic scene construction, high-fidelity embodied simulation, and navigation policy evaluation in visually grounded indoor environments. In the GN0 ecosystem, GN-BAE serves as the navigation foundation model for map-based and map-free policy learning.

Intended Use

This model is intended for research on embodied AI and Vision-and-Language Navigation, especially evaluation with the GN0 / GN-Bench workflow.

Use this checkpoint with the GN0 codebase:

• Code: https://github.com/TeleHuman/GN0
• Project page: https://telehuman-gn0.github.io/
• Paper: https://arxiv.org/abs/2606.03682

GN0 Context

GN0 consists of three connected components:

Component	Description
GN-Matrix	Large-scale 3DGS navigation data with dynamic human avatars.
GN-Bench	Interactive benchmark and simulator for high-fidelity VLN evaluation.
GN-BAE	Navigation foundation model for map-based and map-free policy learning.

This repository hosts the GN-BAE checkpoint referenced by the GN0 model zoo.

Usage

Please follow the installation and evaluation instructions in the GN0 repository.

A typical GN0 workspace places this checkpoint under model_zoo/bae:

GN0
├── data
│   ├── datasets
│   │   └── GN_Matrix
│   │       └── InteriorGS_episode
│   └── scene_datasets
│       └── InteriorGS
├── GN-Bench-Tools
├── model_zoo
│   └── bae
├── VLN_CE
├── run.py
└── eval_bae_InteriorGS.sh

Example InteriorGS evaluation command:

zsh eval_bae_InteriorGS.sh \
  --model-path model_zoo/bae \
  --chunks 1 \
  --procs-per-gpu 1 \
  --save-path tmp/bae_eval

Monitor evaluation progress:

watch -n 1 python analyze_results.py --path tmp/bae_eval

Evaluation

GN0 reports standard VLN metrics from JSON evaluation logs:

Metric	Meaning
TL	Average trajectory length
NE ↓	Navigation error
OS ↑	Oracle success
SR ↑	Success rate
SPL ↑	Success weighted by path length

Please refer to the GN0 repository for dataset preparation, simulator setup, and full evaluation details.

License

This model is released under the Creative Commons Attribution-NonCommercial 4.0 International license (CC BY-NC 4.0).

Citation

If GN0, GN-Bench, or this model is useful for your research, please cite:

@article{li2026gn0,
  title={GN0: Toward a Unified Paradigm for Generation, Evaluation, and Policy Learning in Visual-Language Navigation},
  author={Li, Xinhai and Zhang, Xiaotao and Huang, Yuehao and Dong, Jiankun and Wang, Tianhang and Zhou, Sunyao and Wu, Yunzi and Sun, Chengnuo and Ge, Yunfei and Weng, Qizhen and Zhang, Chi and Bai, Chenjia and Li, Xuelong},
  journal={arXiv preprint arXiv:2606.03682},
  year={2026}
}

Acknowledgements

GN-Bench-Tools is adapted from Habitat-Lab and customized for 3D Gaussian Splatting-based navigation. We thank the Habitat-Lab developers, the InteriorGS authors, and the broader Embodied AI and 3DGS open-source communities.