Spherical Multi-Modal Place Recognition for Heterogeneous Sensor Systems

Overview

In the context of robotics, place recognition is a fundamental problem for autonomous systems. It yields a estimated position of a robot in a prior map given the current observations. We propose an end-to-end multi-modal approach that directly operates on camera images and LiDAR scans without the necessity of a feature extraction. All modalities are projected onto a hypersphere and given as input to a spherical CNN that learns a unique embedding optimized for distinguishing between different places.

Installation

S2Loc was written using PyTorch (http://pytorch.org/) and depends on a few libraries.

• s2cnn: https://github.com/jonas-koehler/s2cnn
• lie_learn: https://github.com/AMLab-Amsterdam/lie_learn
• pynvrtc: https://github.com/NVIDIA/pynvrtc

Submodule references to these repositories can be found in the lib folder

Usage

Clone this repository:

git clone [email protected]:ethz-asl/s2loc.git --recursive

To train a new model the use train.py.

Data Format

path_to_dataset/
    training_anchor_pointclouds/
    training_positive_pointclouds/
    training_negative_pointclouds/
    training_anchor_sph_images/
    training_positive_sph_images/
    training_negative_sph_images/
    anchor-poses.csv
    positive-poses.csv
    negative-poses.csv
    missions.csv

Images need to be projected separately, whereas pointclouds will be projected by the training set provider. Missions are hash ids that are used to separate test and training places. An example training set is provided in the share/ folder.

Reference

Our paper is available at

Bernreiter, Lukas, Lionel Ott, Juan Nieto, Roland Siegwart, and Cesar Cadena. "Spherical Multi-Modal Place Recognition for Heterogeneous Sensor Systems." In 2021 IEEE International Conference on Robotics and Automation (ICRA), pp. 1743-1750. IEEE, 2021. [Link] [ArXiv]

BibTex:

@inproceedings{bernreiter2021spherical,
  title={Spherical Multi-Modal Place Recognition for Heterogeneous Sensor Systems},
  author={Bernreiter, Lukas and Ott, Lionel and Nieto, Juan and Siegwart, Roland and Cadena, Cesar},
  booktitle={2021 IEEE International Conference on Robotics and Automation (ICRA)},
  pages={1743--1750},
  year={2021},
  organization={IEEE}
}