Evaluation of LeGO-LOAM: Lightweight andGround-Optimized Lidar Odometry and Mapping

This is Team 11's final project git repository for NAVARCH/EECS 568, ROB 530: Mobile Robotics. The title of our project is Evaluation of LeGO-LOAM: Lightweight and Ground-Optimized Lidar Odometry and Mapping. Team members include: Bowen Mu, Chien Erh Lin, Haochen Wu, and Weilin Xu.

Abstract

LeGO-LOAM, a lightweight and ground-optimized lidar odometry and mapping method, is able to do six degree-of-freedom pose estimation for real time with ground vehicles. In this project we evaluate the LeGO-LOAM which reduced computational expense while keeping similar accuracy compared to LOAM method using codes provided by RobustFieldAutonomyLab/LeGO-LOAM and laboshinl's version of LOAM. The data used for evaluation is raw data (with Velodyne LiDAR data and IMU) from KITTI Odometry Benchmark Dataset, UTBM Dataset and KAIST Urban Dataset. We compared the mapping and odometry results of these data, and analysis the relative motion and mapping error in this report.

Evaluation Result & Conclusion

The evaluation result on KITTI data set illustrates that position estimation error for odometry is acceptable (around 10 m) and that mapping result can provide concise but adequate information of the environment. By comparing with LOAM, LeGO-LOAM provides better accuracy on mapping and odometry tracking (both rotational and translational), and more efficient since filtering the unreliable point cloud. It is also observed that the loop closure reduces the error in odometry for LeGO-LOAM, and LOAM does not have loop closure ability. However, when the drift is large within a short time (such as sharp turn), the performance of LeGO-LOAM is negatively affected. In addition, from the test on UTBM, LeGO-LOAM also has a limited mapping and tracking ability for the roundabout trajectory.

Getting Started

We recommend you read through the original LOAM paper by Ji Zhang and Sanjiv Singh and the original LeGO-LOAM paper by Tixiao Shan and Brendan Englot. These will give you theoretical understanding of the LOAM and LeGO-LOAM algorithm.

LeGO-LOAM

The LeGO-LOAM repository is forked from RobustFieldAutonomyLab/LeGO-LOAM. We changed it to be able to run kitti dataset.

Prerequisites

• First, install ROS. We used ROS-kinetic on Ubuntu 16.04. Please follow the instruction of ROS installation page.
• gtsam. GTSAM is a library of C++ classes that implement smoothing and mapping (SAM). You can install it by following code.

git clone https://bitbucket.org/gtborg/gtsam.git
cd gtsam/
mkdir build
cd build
cmake ..
make check 
make install

in /usr/local/lib/cmake/GTSAM/GTSAMConfig.cmake, change :17 find_dependency to find_package

Install, Compile, and RUN

To install and compile the code, please clone this respository in src/ folder under catkin workspace.

cd ~/catkin_ws/src/
git https://github.com/chienerh/LeGO-LOAM.git
cd ..
catkin_make -j1

For the first time compiling, add -j1 after catkin_make.

source ./devel/setup.bash
roslaunch lego_loam run.launch  > PATH_TO_TXT_FILE.txt
rosbag play PATH_TO_BAG_FILE.bag --clock --topic /kitti/velo/pointcloud /kitti/oxts/imu

LeGO-LOAM with KITTI Data

• Kitti ran with DHL-64E, so change parameters in utility.h

extern const int N_SCAN = 64;
extern const int Horizon_SCAN = 1800;
extern const float ang_res_x = 0.2;
extern const float ang_res_y = 0.427;
extern const float ang_bottom = 24.9;
extern const int groundScanInd = 50;

• In featureAssociation.cpp line 849 TransformToStart(), float s = 1;
• In featureAssociation.cpp line 1752 and 1758, disable TransformToEnd()
• Sample ros bag files from this link and put it in your own PATH_TO_BAG_FILE folder.
• Currently for our kitti rosbag, lidar topic name is /kitti/velo/pointcloud, and the imu topic name is /kitti/oxts/imu. Change topic names from these two to the topic name in your the rosbag if these are not the correct topic names in run.launch file line 16 and the rosbag play command.

LeGO-LOAM with UTBM Dataset

• According to this repository
• rosbag can be downloaded on their website
• roslaunch lego_loam lego_loam_utbm.launch bag:=PATH_TO_BAG

LOAM

The loam_velodyne repository is forked from laboshinl's version of LOAM. We changed it to be able to run kitti dataset.

Install, Compile, and RUN

cd ~/catkin_ws/src/
git clone https://github.com/chienerh/loam_velodyne
cd ~/catkin_ws
catkin_make -DCMAKE_BUILD_TYPE=Release 
source ~/catkin_ws/devel/setup.bash

roslaunch loam_velodyne loam_velodyne.launch
rosbag play PATH_TO_BAG_FILE.bag 

LOAM with KITTI data

• Rostopic of kitti point cloud is "/kitti/velo/pointcloud", so remap "/multi_scan_points" to "/kitti/velo/pointcloud" in loam_velodyne.launch.
• Set lidar to be "VLP-16" in loam_velodyne.launch.

LOAM with UTBM Dataset

• According to this repository
• rosbag can be downloaded on their website
• install GPS driver
• roslaunch loam_velodyne loam_velodyne_utbm.launch bag:=PATH_TO_BAG

LOAM with KAIST Dataset

• Use File Player to publish KAIST data into ros message.
• Rostopic of KAIST dataset point cloud topic is "/ns1/velodyne_points", so remap "/multi_scan_points" to "/ns1/velodyne_points" in loam_velodyne.launch.

Result

Kitti Dataset

LeGO-LOAM

LOAM

UTBM Dataset

APPENDIX

How to download and transform kitti to rosbag

• Download raw sequences synced+rectified data and calibration data from The KITTI Vision Benchmark Suite.
• kitti2bag is used to transform kitti raw data to rosbag.

The sequences that provide ground truth are shown in the below table. The odometry eval kit includes description.

The following table lists the name, start and end frame of each sequence that has been used to extract the visual odometry / SLAM training set

Seq	Sequence name	Start	End	Category	Size
00	2011_10_03_drive_0027	000000	004540	Residential	17.6 GB
01	2011_10_03_drive_0042	000000	001100	Road	4.5 GB
02	2011_10_03_drive_0034	000000	004660	Residential	18.0 GB
03	2011_09_26_drive_0067	000000	000800
04	2011_09_30_drive_0016	000000	000270	Road	1.1 GB
05	2011_09_30_drive_0018	000000	002760	Residential	10.7 GB
06	2011_09_30_drive_0020	000000	001100	Residential	4.3 GB
07	2011_09_30_drive_0027	000000	001100	Residential	4.3 GB
08	2011_09_30_drive_0028	001100	005170	Residential	20.0 GB
09	2011_09_30_drive_0033	000000	001590	Residential	6.2 GB
10	2011_09_30_drive_0034	000000	001200	Residential	4.8 GB

How to save the odometry result

• in launch file, add

  <!-- rosbag arg -->
  <arg name="ros_bag_name" default="simulation_res"/>

  <node pkg="rosbag" type="record" name="record" args="record -O /home/USER_NAME/catkin_ws/src/loam_velodyne/result/bag/$(arg ros_bag_name).bag
                                                	/integrated_to_init
							/groundtruth_pose/pose"/>/

• to save odometry result, in src/lib/TransformMaintence.cpp line 79, add

   printf("%f, %f, %f\n", transformMapped()[3], transformMapped()[4], transformMapped()[5]);

Then run launch file with this command outputing file to a csv

roslaunch loam_velodyne loam_velodyne.launch > PATH_TO_FILE.csv

The odmetry result in saved PATH_TO_TXT_FILE.txt and also in ./src/LeGO-LOAM/result/bag/. This .bag file can transformed to .csv file by:

python ./src/LeGO-LOAM/result/process_rosbag.py

Afterwards, you can use ./src/result_plot.m in the repo to plot and compare PATH_TO_TXT_FILE.txt and the ground truth odemetry txt file. You need to change variables res_file and groundTruthFile in this MATLAB to the correct name of PATH_TO_TXT_FILE.txt and ground truth odemetry txt.