SLAM_Demos

This project, built on ROS, provides some basic demos of SLAM. The mobile robot Pioneer 3-DX is employed in the project, integrating with two different depth sensors, namely, a laser radar from HOKUYO and an RGB-D camera Kinect. Users are able to conduct SLAM by relying on either of these two sensors and compare the maps generated by them.

Testing Platform

The code was tested recently with the following configurations and worked well. I believe it should still work with the previous configurations (see below) since only trivial changes were made this time (but perhaps no one will follow that outdated configuration now).

Ubuntu: 20.04
ROS: noetic
Gazebo: 11.11

Last Test

Ubuntu: 16.04

ROS: kinetic

Gazebo: 8.6

I believe Gazebo 7.0, the default version within ROS kinetic, should also be fine to run this project because it doesn't depend on any version-specific libraries from Gazebo.

Reminder: If someone finds the world file cannot be loaded successfully, please check if you have downloaded and placed correctly the model library for Gazebo. If not, please download it by

git clone git@github.com:osrf/gazebo_models.git

and place all of them at ~/.gazebo/models.

SLAM by Lidar

The following command will open Gazebo and load the robot model in Gazebo. The launching option kinect:=false can be omitted since it is the default value.

roslaunch p3dx_gazebo p3dx_gazebo.launch kinect:=false

If everything is fine, we can see the robot in Gazebo as shown in the picture below.

Then, the following command will launch a ROS node gmapping together with RViz. The launching option kinetic:=false is default and can be omitted, too.

roslaunch p3dx_slam gmapping_rviz.launch kinect:=false

After that, the generated map is displayed in RViz.

We can then drive the robot in the simulated workspace by sending velocity commands via the ROS node teleop_twist_keyboard. A map will be established while the robot is moving around. An example of the generated map is shown below. You can find the picture of this map at p3dx_slam/maps/map_laser.pgm together with a configuration file map_laser.yaml, which can be used to achieve robot navigation after further development.

SLAM by Kinect

It is quite simple to switch from "SLAM by Lidar" to "SLAM by Kinect". The only thing to do is specify the option kinect:=true when launching the simulation. Similar with the former example, the following commands will open Gazebo and RViz, and load the robot which is equipped with a Kinect to the simulated workspace.

roslaunch p3dx_gazebo p3dx_gazebo.launch kinect:=true
roslaunch p3dx_slam gmapping_rviz.launch kinect:=true

We can observe that a Kinect has been installed at top of the robot.

By launching the ROS node teleop_twist_keyboard and moving the robot around (better in a very slow speed this time), a map will be generated as before. An example map is saved at p3dx_slam/maps, which is also shown below. By comparing the results of using the lidar and using the Kinect, it is obvious that the lidar does a better job in SLAM (considerable slippage and distortion can be observed in the map generated by the Kinect).

A Third Mode

As seen above, it is more expected to conduct SLAM by using the lidar. However, the Kinect gives a kind of more realistic way to sense the surroundings. This project allows users to build up maps by the lidar while observe surroundings through the Kinect. Have a try with the following commands.

roslaunch p3dx_gazebo p3dx_gazebo.launch kinect:=true
roslaunch p3dx_slam gmapping_rviz.launch kinect:=false

In the Displays panel of RViz, click Add, select the tag PointCloud2 and then set the topic to be /kinect/depth/points. If everything is fine, you will have a similar case as shown below, where SLAM is conducted by using the lidar and the surroundings are visualized by the Kinect.

Known Issues

Luckily there are few known issues ...

However, there is indeed one operation users should avoid when running simulations, that is, not launch the simulation like this.

roslaunch p3dx_gazebo p3dx_gazebo.launch kinect:=false
roslaunch p3dx_slam gmapping_rviz.launch kinect:=true

You will meet problems of building up maps by doing so (although all the components seem to work well) since the ROS node gmapping, which is responsible for map construction, expects data from the topic /kinect/scan but the robot is not equipped with a Kinect currently.

Check the list of all the ROS topics.