No RL No Simulation (NRNS)

Official implementation of the NRNS paper: No RL, No Simulation: Learning to Navigate without Navigating

NRNS is a heriarchical modular approach to image goal navigation that uses a topological map and distance estimator to navigate and self-localize. Distance function and target prediction function are learnt over passive video trajectories gathered from Mp3D and Gibson.

[project website]

Setup

This project is developed with Python 3.6. If you are using miniconda or anaconda, you can create an environment:

conda create -n nrns python3.6
conda activate nrns

Install Habitat and Other Dependencies

NRNS makes extensive use of the Habitat Simulator and Habitat-Lab developed by FAIR. You will first need to install both Habitat-Sim and Habitat-Lab.

Please find the instructions to install habitat here

If you are using conda, Habitat-Sim can easily be installed with

conda install -c aihabitat -c conda-forge habitat-sim headless

We recommend downloading the test scenes and running the example script as described here to ensure the installation of Habitat-Sim and Habitat-Lab was successful. Now you can clone this repository and install the rest of the dependencies:

git clone git@github.com:meera1hahn/NRNS.git
cd NRNS
python -m pip install -r requirements.txt
python download_aux.py

Download Scene Data

Like Habitat-Lab, we expect a data folder (or symlink) with a particular structure in the top-level directory of this project. Running the download_aux.py script will download the pretrained models but you will still need to download the scene data. We evaluate our agents on Matterport3D (MP3D) and Gibson scene reconstructions. Instructions on how to download RealEstate10k can be found here.

Image-Nav Test Episodes

The image-nav test epsiodes used in this paper for MP3D and Gibson can be found here. These were used to test all baselines and NRNS.

Matterport3D

The official Matterport3D download script (download_mp.py) can be accessed by following the "Dataset Download" instructions on their project webpage. The scene data can then be downloaded this way:

# requires running with python 2.7
python download_mp.py --task habitat -o data/scene_datasets/mp3d/

Extract this data to data/scene_datasets/mp3d such that it has the form data/scene_datasets/mp3d/{scene}/{scene}.glb. There should be 90 total scenes. We follow the standard train/val/test splits.

Gibson

The official Gibson dataset can be accessed on their project webpage. Please follow the link to download the Habitat Simulator compatible data. The link will first take you to the license agreement and then to the data. We follow the standard train/val/test splits.

Running pre-trained models

Look at the run scripts in src/image_nav/run_scripts/ for examples of how to run the model.

Difficulty settings options are: easy, medium, hard

Path Type setting options are: straight, curved

To run NRNS on gibson without noise for example on the straight setting with a medium difficulty

cd src/image_nav/
python -W ignore run.py \
    --dataset 'gibson' \
    --path_type 'straight' \
    --difficulty 'medium' \

Scores

Expected scores for NRNS from running this code without noise

Gibson

Path Type	Success	SPL
Easy - Straight	0.7010	0.6271
Medium - Straight	0.5070	0.4153
Hard - Straight	0.2122	0.1547
Easy - Curved	0.3140	0.1070
Medium - Curved	0.2200	0.0829
Hard - Curved	0.1190	0.0544

Matterport

Path Type	Success	SPL
Easy - Straight	0.5690	0.4926
Medium - Straight	0.3370	0.2717
Hard - Straight	0.1960	0.1444
Easy - Curved	0.2310	0.1086
Medium - Curved	0.1510	0.0738
Hard - Curved	0.0840	0.0414

Citing

If you use NRNS in your research, please cite the following paper:

@inproceedings{hahn_nrns_2021,
  title={No RL, No Simulation: Learning to Navigate without Navigating},
  author={Meera Hahn and Devendra Chaplot and Mustafa Mukadam and James M. Rehg and Shubham Tulsiani and Abhinav Gupta},
  booktitle={Neurips},
  year={2021}
 }