F1/10 @ CTU (ROS 2 port) with dynamic obstacle avoidance

This is an implementation of dynamic obstacle avoidance to CTU's f1tenth codebase. This code is based on an implementation from f1tenth-rewrite (at https://github.com/CTU-F1T/f1tenth-rewrite). The demonstration and presentation of this project is in this folder.

Localization

Cartographer slam was selected for localization of the car on the track. Modified configuration files can be found in this folder. Two sets of measurements were done to evaluate the performance of the localization. Detailed description of the measuring procedure and the results can be found in this file.

Implementation

Map creation

The implementation of main algorithm is written in 'follow_center_line' package. It uses the map from 'cartographer_slam' and combines it with data from LIDAR to get the information about dynamic obstacles.

The points from LIDAR are binary dilated by the factor of 6, then combined with the map and the entire map is subsequently dilated by the factor of 5. The subsequent skeletonization for path generation uses method 'zha' which is a fast parallel algorithm for thinning digital patherns.

Path follow

The BFS algorithm finds the nearest point on the path from the current position of the car and DFS follows the path to the depth 40 (map cells) and checks whether the path still exists there. This is to avoid false branches that can occur after skeletonization. The nearest point found with BFS is then taken as a look ahead point if DFS is succesful. The car is stopped if DFS is not succesful.

Unsuccessful methods

The following methods for skeletonization were not succesful: 'lee', 'medial_axis', because they create more side branches and run longer than the selected one.

We also tried EDF algorithm for path generation but it did not solve the problem with several branches therefore skeletonization was use instead.

Building

# Clone the directory
$ git clone https://github.com/CTU-F1T/pvt_2023
# Get vesc and messages submodule
$ git submodule update --init --recusrsive
# Colcon build
$ colcon build --symlink-install --mixin compile-commands --packages-ignore vesc_ackermann

Running

It is necessary to create the map first, for just mapping, use the following commands:

$ ros2 launch auto auto.ftg.launch.py config:=tx2-auto-usa.yaml
$ ros2 launch cartographer_slam mapping.launch.py

Cartographer searches for the saved map in a specific folder, copy it there first:

$ auto save_slam track
$ cp track.pbstream /home/nvidia/ros2_ws/install/storage/share/storage/stage/world

Then run all required nodes:

$ ros2 launch auto auto.ftg.launch.py config:=tx2-auto-usa.yaml
$ ros2 launch cartographer_slam localization.launch.py map:=track.pbstream
$ ros2 launch follow_center_line follow_line.launch.py

After publishing 'False' on topic '/eStop' the car should start moving.