Learning-Aided 3D Mapping

A suite of algorithms for learning-aided mapping. Includes implementations of Gaussian process regression and Bayesian generalized kernel inference for occupancy prediction using test-data octrees. This framework also contains the components necessary to run OctoMap as a baseline.

Overview

This implementation as it stands now is primarily intended to enable replication of these methods over a few datasets. In addition to the implementation of relevant learning algorithms and data structures, we provide two sets of range data (sim_structured and sim_unstructured) collected in Gazebo for demonstration. Parameters of the sensors and environments are set in the relevant yaml files contained in the config/datasets directory, while configuration of parameters for the mapping methods can be found in config/methods.

Getting Started

Dependencies

We tested LA3DM with ROS Kinetic, but it also works with ROS Indigo, just ensure you have the correct dependencies by running:

$ sudo apt-get install ros-kinetic-octomap*

if you're using ROS Kinetic, or:

$ sudo apt-get install ros-indigo-octomap*

if you're using Indigo.

Building with catkin

The repository is set up to work with catkin, so to get started you can clone the repository into your catkin workspace src folder and compile with catkin_make:

my_catkin_workspace/src$ git clone https://github.com/RobustFieldAutonomyLab/la3dm
my_catkin_workspace/src$ cd ..
my_catkin_workspace$ catkin_make
my_catkin_workspace$ source ~/my_catkin_workspace/devel/setup.bash

Running the Demo

To run the demo on the sim_structured environment, simply run:

$ roslaunch la3dm la3dm_static.launch

which by default will run using the BGKOctoMap method. If you want to try a different method or dataset, simply pass the name of the method or dataset as a parameter. For example, if you want to run standard OctoMap on the sim_unstructured map, you would run:

$ roslaunch la3dm la3dm_static.launch method:=octomap dataset:=sim_unstructured

Relevant Publications

If you found this code useful, please cite the following:

Learning-Aided 3-D Occupancy Mapping with Bayesian Generalized Kernel Inference (PDF)

@article{Doherty2019,
  doi = {10.1109/tro.2019.2912487},
  url = {https://doi.org/10.1109/tro.2019.2912487},
  year = {2019},
  publisher = {Institute of Electrical and Electronics Engineers ({IEEE})},
  pages = {1--14},
  author = {Kevin Doherty and Tixiao Shan and Jinkun Wang and Brendan Englot},
  title = {Learning-Aided 3-D Occupancy Mapping With Bayesian Generalized Kernel Inference},
  journal = {{IEEE} Transactions on Robotics}
}

Fast, accurate gaussian process occupancy maps via test-data octrees and nested Bayesian fusion (PDF)

@INPROCEEDINGS{JWang-ICRA-16,
author={J. Wang and B. Englot},
booktitle={2016 IEEE International Conference on Robotics and Automation (ICRA)},
title={Fast, accurate gaussian process occupancy maps via test-data octrees and nested Bayesian fusion},
year={2016},
pages={1003-1010},
month={May},
}

Bayesian Generalized Kernel Inference for Occupancy Map Prediction (PDF)

@INPROCEEDINGS{KDoherty-ICRA-17,
author={K. Doherty and J. Wang, and B. Englot},
booktitle={2017 IEEE International Conference on Robotics and Automation (ICRA)},
title={Bayesian Generalized Kernel Inference for Occupancy Map Prediction},
year={2017},
month={May},
}

Contributors

Jinkun Wang and Kevin Doherty, Robust Field Autonomy Lab (RFAL), Stevens Institute of Technology.