Exploiting Proximity-Aware Tasks for Embodied Social Navigation

Enrico Cancelli, Tommaso Campari, Luciano Serafini, Angel X. Chang, Lamberto Ballan

Accepted at ICCV 2023

Short version: [CVPR EAI workshop paper]

This repository contains the implementation and episode dataset from the paper.

Warning: It is still WIP. Feel free to open an issue if you encounter some problems or have some questions.

Setup

Setting up conda

conda create -n socialnav python=3.6
conda activate socialnav

pip install cython

Installing the project

Our code is based on habitat-lab 0.1.7 (please refer to the original repo). Clone this repo and install it.

cd habitat-lab
pip install -e .

(plus habitat-sim 0.2.1, see original repo for building options)

conda install habitat-sim=0.2.1 withbullet headless -c conda-forge -c aihabitat

Also install both habitat-lab and habitat-sim requirements

Getting people's meshes

People's meshes are taken from the original challenge data. To get them, clone the original repository using the following script:

git clone https://github.com/StanfordVL/iGibsonChallenge2021.git

cd ../iGibsonChallenge2021
./download.sh

mkdir ../iGibson/gibson2/data
mv gibson_challenge_data_2021/* ../iGibson/gibson2/data

After download, substitute the PATH_TO_PEOPLE_MESHES variable in the habitat-lab/habitat/sims/igibson_challenge/social_nav.py with the correct path (will be made a parameter in future).

HM3D-S episode dataset

The dataset is in the habitat-lab/data/dataset/pointnav/hm3d folder. To use it add this parameter to your main config file:

BASE_TASK_CONFIG_PATH: configs/datasets/pointnav/hm3d.yaml

Examples

TODO: this section is work in progress. For now you can use the following command with the associated config for the baseline:

python habitat_baselines/run.py --exp-config ../baseline.yaml --run-type eval (or train) --name example

and this for te full model:

python habitat_baselines/run.py --exp-config ../full.yaml --run-type eval (or train) --name example Please consult the original repository for details about the config system and habitat-lab and habitat_baselines parameters.