GridMM: Grid Memory Map for Vision-and-Language Navigation

Zihan Wang, Xiangyang Li, Jiahao Yang, Yeqi Liu and Shuqiang Jiang

This repository is the official implementation of GridMM: Grid Memory Map for Vision-and-Language Navigation.

Vision-and-language navigation (VLN) enables the agent to navigate to a remote location following the natural language instruction in 3D environments. To represent the previously visited environment, most approaches for VLN implement memory using recurrent states, topological maps, or top-down semantic maps. In contrast to these approaches, we build the top-down egocentric and dynamically growing Grid Memory Map (i.e., GridMM) to structure the visited environment. From a global perspective, historical observations are projected into a unified grid map in a top-down view, which can better represent the spatial relations of the environment. From a local perspective, we further propose an instruction relevance aggregation method to capture fine-grained visual clues in each grid region. Extensive experiments are conducted on both the REVERIE, R2R, SOON datasets in the discrete environments, and the R2R-CE dataset in the continuous environments, showing the superiority of our proposed method.

Requirements

1. Install Matterport3D simulator for R2R, REVERIE and SOON: follow instructions here.

export PYTHONPATH=Matterport3DSimulator/build:$PYTHONPATH

2. Install requirements:

conda create --name GridMM python=3.8.0
conda activate GridMM
pip install -r requirements.txt

3. Download annotations, preprocessed features, and trained models from Baidu Netdisk.

4. Install Habitat simulator for R2R-CE: follow instructions here and here.

Pretraining

Combine behavior cloning and auxiliary proxy tasks in pretraining:

cd pretrain_src
bash run_r2r.sh # (run_reverie.sh, run_soon.sh)

Fine-tuning & Evaluation for R2R, REVERIE and SOON

Use pseudo interative demonstrator to fine-tune the model:

cd map_nav_src
bash scripts/run_r2r.sh # (run_reverie.sh, run_soon.sh)

Fine-tuning & Evaluation for R2R-CE

Use pseudo interative demonstrator to fine-tune the model:

cd VLN_CE
bash run_GridMap.bash  # Currently, this code only supports evaluation with a single GPU.

Citation

@inproceedings{wang2023gridmm,
  title={Gridmm: Grid memory map for vision-and-language navigation},
  author={Wang, Zihan and Li, Xiangyang and Yang, Jiahao and Liu, Yeqi and Jiang, Shuqiang},
  booktitle={Proceedings of the IEEE/CVF International Conference on Computer Vision},
  pages={15625--15636},
  year={2023}
}

Acknowledgments

Our code is based on VLN-DUET and CWP. Thanks for their great works!