LOL: Lidar-only Odometry and Localization in 3D point cloud maps

Supplementary material for our ICRA 2020 paper

Abstract - In this paper we deal with the problem of odom- etry and localization for Lidar-equipped vehicles driving in urban environments, where a premade target map exists to localize against. In our problem formulation, to correct the accumulated drift of the Lidar-only odometry we apply a place recognition method to detect geometrically similar locations between the online 3D point cloud and the a priori offline map. In the proposed system, we integrate a state-of-the-art Lidar- only odometry algorithm with a recently proposed 3D point segment matching method by complementing their advantages. Also, we propose additional enhancements in order to reduce the number of false matches between the online point cloud and the target map, and to refine the position estimation error whenever a good match is detected. We demonstrate the utility of the proposed LOL system on several Kitti datasets of different lengths and environments, where the relocalization accuracy and the precision of the vehicle’s trajectory were significantly improved in every case, while still being able to maintain real-time performance.

arXiv | Video

Advances to the state-of-the-art:

We present a novel Lidar-only odometer and Localization system by integrating and complementing the advantages of two state of the algorithms.
We propose a set of enhancements: (i) a RANSAC-based geometrical verification to reduce the number of false matches between the online point cloud and the offline map; and (ii) a fine-grained ICP alignment to refine the relocalization accuracy whenever a good match is detected.
Here we publicly release the source code of the proposed system with supplementary prepared datasets to test.

For any code-related or other questions open an issue here. If you found this work helpful for your research, please cite our paper:

@inproceedings{lol_3dlocalization, 
	title={{LOL: Lidar-only Odometry and Localization in 3D point cloud maps}}, 
	author={Rozenberszki, Dávid and Majdik, András}, 
	booktitle={2020 IEEE International Conference on Robotics and Automation (ICRA)}, 
	year={2020},
        organization={IEEE} 
}

Installation

Ubuntu 64-bit 16.04. ROS Kinetic. ROS Installation

Install the dependencies:

$ sudo apt-get update
$ sudo apt-get upgrade
$ sudo apt-get install python-wstool python-catkin-pkg doxygen python3-pip python3-dev python-virtualenv dh-autoreconf python-catkin-tools

Now create the Catkin Environment:

$ mkdir -p ~/Catkin/LOL/src
$ cd ~/Catkin/LOL
$ catkin init
$ catkin config --merge-devel
$ catkin config --cmake-args -DCMAKE_BUILD_TYPE=Release

And clone the project:

$ git clone https://github.com/RozDavid/LOL.git
$ wstool init
$ wstool merge segmap/dependencies.rosinstall
$ wstool update

The CNN descriptors were made in Tensorflow, for compiling the whole package and using the localization function with learning based descriptors one needs to install for Ubuntu 16.04

Finally build the packages with:

$ cd ~/Catkin/LOL
$ catkin build tensorflow_ros_cpp
$ catkin build segmapper
$ catkin build loam_velodyne

###Rosbag Examples

Download the provided map resources to your machine from here and save them anywhere in your machine. The Rosbags for the examples could be downloaded from the original Kitti dataset website, you just need to strip other sensor measurement and /tf topic from it to run correctly.

Then modify the folowing launch and yaml and set the path for downloaded dataset files

$ ~/Catkin/LOL/segmap/segmapper/launch/kitti/cnn_kitti_loam_segmap.yaml
$ ~/Catkin/LOL/segmap/segmapper/launch/kitti/cnn_loam_segmap.launch
$ ~/Catkin/LOL/segmap/segmapper/launch/kitti/kitti_loam_segmap.launch
$ ~/Catkin/LOL/segmap/segmapper/launch/kitti/kitti_localization.yaml

Running the Examples

Run Localization with Eigen features

roslaunch segmapper kitti_loam_segmap.launch

Run Localization with CNN features

roslaunch segmapper cnn_loam_segmam.launch

Acknowledgement

This source code and the resulting paper is highly dependent and mostly based on two amazing state-of-the art algorithms. The Odometry is calculated by the LOAM, while the segmentation, feature detection and matching is based on the SegMap algorithm. Without these works this paper wouldn't be able to exist.

The research reported in this paper was supported by the Hungarian Scientific Research Fund (No. NKFIH OTKA KH-126513) and by the project: Exploring the Mathematical Foundations of Artificial Intelligence 2018-1.2.1-NKP-00008.

RozDavid / LOL