Improving accuracy of Hector SLAM using DBSCAN

DBSCAN을 활용하여 Hector SLAM과정에서의 정확도 향상 연구

Hardware Info

OS: Ubuntu 20.04(ROS1 Noetic)
HW: RPLiDAR A3

Installation

0. Project folder

cd ~/catkin_ws
mkdir -p src
cd src

1. RPLiDAR A3 SDK

git clone https://github.com/Slamtec/rplidar_ros.git

2. Hector SLAM

git clone https://github.com/tu-darmstadt-ros-pkg/hector_slam.git

modify /hector_mapping/launch/mapping_default.launch line 5, 6 into:

<arg name="base_frame" default="base_link"/>
<arg name="odom_frame" default="base_link"/>

3. This package

git clone https://github.com/Kimdoodle/lidar_slam.git

4. build

cd ~/catkin_ws
catkin_make

Run

Option 1. Run both lidar and slam

roslaunch lidar_slam hector_lidar.launch

Option 2. Run seperately

roslaunch rplidar_ros rplidar_a3.launch
roslaunch lidar_slam hector_only.launch

Parameters

Change params in /launch/hectormapping.launch

<param name="eps_ratio" value="5"/>
<param name="minpts" value="15"/>
<param name="remains" value="0.5"/>

1. eps_ratio(0~100) stands for the eps value in DBSCAN, calculated as n% of the Euclidean distance in each scan.
2. minpts stands for the minimum number of points required to form a cluster.
3. remains(0.0 ~ 1.0) stands for the remaining clusters, which sorts all clusters by the number of points and retains only n% of them. Other clusters' labels are turned into -1 (outliers).