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.
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
.
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.
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
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
If you found this code useful, please cite the following:
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},
}
Jinkun Wang and Kevin Doherty, Robust Field Autonomy Lab (RFAL), Stevens Institute of Technology.