VL-LN-Bench-basemodel

VL-LN-Bench basemodel

This repository contains the base model for the paper VL-LN Bench: Towards Long-horizon Goal-oriented Navigation with Active Dialogs.

Model Description

VL-LN Bench is the first benchmark for Interactive Instance Goal Navigation (IIGN), where an embodied agent must locate a specific object instance in a realistic 3D home while engaging in free-form natural-language dialogue. It also provides an automated data-collection pipeline that generates large-scale training data for learning interactive navigation behaviors. Using this dataset, we train an IIGN base model that shares the same architecture as InternVLA-N1.

The resulting model demonstrates baseline competence on IIGN: it can search for a specific instance in previously unseen environments. During exploration, the agent can either move by predicting a pixel-goal waypoint or ask a question to reduce ambiguity and improve task success and efficiency.

Resources

Usage

For inference and evaluation, please refer to the VL-LN-Bench repository.

Citation

If you find our work helpful, please cite:

@misc{huang2025vllnbenchlonghorizongoaloriented,
      title={VL-LN Bench: Towards Long-horizon Goal-oriented Navigation with Active Dialogs}, 
      author={Wensi Huang and Shaohao Zhu and Meng Wei and Jinming Xu and Xihui Liu and Hanqing Wang and Tai Wang and Feng Zhao and Jiangmiao Pang},
      year={2025},
      eprint={2512.22342},
      archivePrefix={arXiv},
      primaryClass={cs.RO},
      url={https://arxiv.org/abs/2512.22342}, 
}