GPU Slam

A CUDA implementation of slam algorithm.

Usage

This project is built with ROS Melodic on Ubuntu 18.04.1. However, other versions should work as well.

Preparing simulation environment

In the first terminal tab:

sudo apt-get install ros-melodic-turtlebot3-gazebo
sudo apt-get install ros-melodic-turtlebot3-description
sudo apt-get install ros-melodic-turtlebot3-teleop

export TURTLEBOT3_MODEL=waffle_pi
roslaunch turtlebot3_gazebo turtlebot3_world.launch
roslaunch turtlebot3_gazebo turtlebot3_house.launch

In the second terminal tab:

export TURTLEBOT3_MODEL=waffle_pi
roslaunch turtlebot3_teleop turtlebot3_teleop_key.launch

WARNING: Do not move or send any velocity command to the robot until you see the following line in the gpu-slam terminal: Initializing: Done. Ready to move

Launching gpu slam

roslaunch gpu-slam gpu-slam.launch

Launching gpu slam

export TURTLEBOT3_MODEL=waffle_pi
roslaunch turtlebot3_slam turtlebot3_slam.launch

Result

Parameters

Parameter	Default	Explanation
map_frame	map	Name of map frame
odom_frame	odom	Name of odometry frame
urbase_frame	urbase_link	Name of un-rotated base frame. This is the base frame of local map
base_frame	base_footprint	Name of base frame
laser_frame	base_scan	Name of laser frame
laser_topic	scan	Name of laser topic to subscribe
particle_topic	particles	Name of topic where particles are published
map_to_odom_publish_period	0.1s	Map to odometry publish interval
map_to_urbase_pl_period	0.1s	Map to un-rotated base frame publish interval
global_map_publish_period	1s	Global map publish interval
local_map_publish_period	0.5s	Local map publish interval
particle_pl_period	0.5s	Particle publish interval
particle_count	5 particles	Number of particles used in the filter
initialization_scan	30 scans	Number of very first scan used to construct initial global map. Do not move or send velocity command to the robot in this stage
x_x	0.01 m	Error factor in x axis caused by x movement
x_y	0.01 m	Error factor in x axis caused by y movement
x_r	0.01 m	Error factor in x axis caused by r movement
y_x	0.01 m	Error factor in y axis caused by x movement
y_y	0.01 m	Error factor in y axis caused by y movement
y_r	0.01 m	Error factor in y axis caused by r movement
r_x	0.01 m	Error factor in r axis caused by x movement
r_y	0.01 m	Error factor in r axis caused by y movement
r_r	0.01 m	Error factor in r axis caused by r movement
min_range	0.1 m	Laser min range for a valid range
max_range	3.5 m	Laser max range for a valid range
corr_iter_thresh_num	5 iterations	Number of iterations in correction step
kernel_size_m	0.1 m	Region around ending point to find correspondance
neff_thresh	0.5	Do resample when neff < neff_thresh
dx_resample_thresh	1 m	Do resample when accumulated movement in x axis exceed this threshold
dy_resample_thresh	1 m	Do resample when accumulated movement in y axis exceed this threshold
dt_resample_thresh	1 m	Do resample when accumulated movement in theta axis exceed this threshold
initial_global_map_width	20 m	Initial global map width in meter
initial_global_map_height	20 m	Initial global map height in meter
initial_local_map_width	10 m	Initial local map width in meter
initial_local_map_height	10 m	Initial local map height in meter
m_per_cell	0.05 m	Size of a square cell in meter
true_positive	0.9	Probability of occupied cell is measured to be occupied
true_negative	0.5	Probability of un-occupied cell is measured to be un-occupied
w_thresh	0.3	Only update map when weight of particle larger than this threshold
dx_map_update_thresh	0.02 m	Do map update only when movement in x axis exceed this threshold
dy_map_update_thresh	0.02 m	Do map update only when movement in y axis exceed this threshold
dt_map_update_thresh	0.02 m	Do map update only when movement in theta axis exceed this threshold

License & copyright

@ Van Tuan Ngo, Ho Chi Minh City University of Technology

Licensed under the MIT license