perception-localization-mapping-projects

ROS Noetic | mobile robot | odometry | slam_toolbox | C++

This repository contains the assignment description and my team's solutions for the two projects of the course in
Perception Localization and Mapping.

The pdf ROS_basics contains my post-lecture notes on the laboratory sessions.

Important

Even if in the pdf assignment there is a Google Drive link to the project data,
unfortunately that Drive is updated to the most recent academic year assignment (not to the one described in this Repo, 2023).
Refer to my Drive link to download the bags used in this project.

first_project (odometry)

Note

bags available at this Drive. The two files P1_first.bag and P1_second.bag refer to two navigation recording session.

We were asked to compute robot odometry given a ROS Bag file, containing the recording of a navigation session for an autonomous shuttle at Politecnico di Milano. The Bag file contains information about the wheel encoder and a 2D LIDAR point cloud for visualization (in RViz).

The main functions of odom_node are:

• Listen to the encoder information, when information arrives it updates odometry using the bicycle kinematic model for the vehicle.
• when encoder data arrives, using a TF broadcaster, the new estimated position is published
• A standard odometry message and a custom message are published (also as encoder callback).

first_project package provide also:

• A Service reset_odom, to reset odometry data (x, y, theta).
• A launch file that runs odom_node, initialize sim_time and odometric variables. Run four static TF publishers for the vehicle wheels' position with respect to base_link.

Run the code:

After installing first_project package (catkin_make on the workspace), as with any other ROS package (remember to source):

First of all, remember the roscore!

# Terminal 0:
roscore

The bag file has to be played in a separate terminal to test the code. Otherwise, this can be tested by publishing "fake" encoder data in the proper topic "speed_steer", with data of the expected type.

# Terminal 1:
rosbag play <robot.bag>

#Or the alternative way to publish encoder data if bag not available

# Terminal 2:
roslaunch first_project odom_launch.launch 

To visualize TF and LIDAR signals, RViz has to be run in another Terminal:

# Terminal 3:
rosrun rviz rviz

Important

In RViz add by topic all the useful information to look at (then it is possible to save a custom RViz config to run automatically). We were not very skilled with ROS, the best practice is to start RViz in the launch itself and define a ".rviz" configuration file.

For further details on the task, refer to first_project assignment
(being the first project with ROS, the structure of the node is not optimal)

second_project (Mapping, SLAM, and Waypoint Navigation)

Note

Bags available at this Drive. The two files P2_first.bag and P2_second.bag refer to two navigation recording sessions.
To launch mapping, upload the bag files on /second_project/bags, or change the path to the bag file in mapping.launch

Given a ROS Bag file with the recording of a teleoperation of the robot above inside the Lab, we have to create a map from scanner data, denoise the map manually, and then set up autonomous navigation on that map. For this project, a lightweight robot simulator is used, called stage (The best practice for simulation is to use Gazebo or other simulators ... I never use stage again to be honest). This simulator is used for a simple kinematic and sensor simulation, but the Navigation stack used in this project is obviously compatible with Gazebo.

The Bag file contains:

• Data from single plane scanner in "/scan"
• Odometry data in "/t265/odom"
• 3d Laser data in "/velodyne_points"

The structure of second_project package is as follows:

• Nodes:
  • tf_publisher:
    Convert odometry data in "/t265/odom" to TF data (necessary conversion for SLAM performed by the navigation toolbox)
  • navigation:
    Manage the waypoint reading from csv file, and the navigation MoveBaseAction requests. (An alternative way to write the same node is navigation(with classes)
• Launch:

  • mapping:
    Start the rosbag (we have two bags available to perform the mapping). Perform the mapping procedure from sensor data, which is necessary before navigation. 3d laser Point cloud (from velodyne) is converted into 2d data by pointcloud_to_laserscan node (with proper parameter initialization).

    The mapping output is then denoised manually, to remove sensor noise that can affect navigation.

  • navigation:
    Launch our navigation node (for waypoint management) and set all the parameters for the navigation toolbox.

Run the code:

• Mapping:
  Remember to upload the bag files as defined above. As usual, start with the roscore, then:

  # Terminal 1:
  roslaunch second_project mapping.launch

  This will start Rviz with the visualization of the online mapping procedure. In Rviz both the 2D laser scan and the 3D sliced as 2D are visible. At the end of the mapping procedure, it is possible to save the map by:

  # Terminal 2:
  rosrun map_saver map_saver

  In the folder where this is called, we save the mapped environment as .pgm and .yaml files. For the project, we save the map created from both P2_first and P2_second recordings, with the map of 2D and 3D laser scan, those are available in /map_raw. Then, the map used for navigation is denoised by hand, to avoid unexpected obstacles that affect navigation.

• Waypoint Navigation:
  The map used for navigation is selected in navigation.launch, as the denoised map in /stage/nav_map.yaml.

  As usual, start with the roscore, then:

  # Terminal 1:
  roslaunch second_project navigation.launch

  This will run both RViz for visualization (properly configured), and the stage simulator.
  navigation node will also run, and the robot will move to the waypoints specified in waypoints.csv. You can define different meaningful waypoints or change some navigation parameters to see how navigation behavior changes.

For further details on the task, refer to second_project assignment.
While, for details on the ROS Navigation stack, you can refer to LAB Slides