This was a project for the course "Technik autonomer Systeme" during my electrical and computational engineering master (WS 16/17) at the Technical University Munich.

This programm will make the car follow a monochrome ball.

It is splitted in three parts:

1. The class find_ball which filters the images of the camera for the ball using OpenCV.
2. The ros-node navigation_goals_node which uses the class find_ball to get information about the ball as well as has the controllers to determine the angular and linear speed and publishs this to the car.
3. The programm calibrate_params which can be used to create a threshold_matrix.xml for any ball.

Try to put the camera below the bumper for best result:

• camera
• monochrome ball
• drivable car
• wii-controller

calibrate parameters for monochrome ball with following command:

rosrun follow_ball calibrate_params

This will need a monitor to display the results. Use the controller such that the ball is white in every picture. In the best case scenario the final image only contains the ball. This doesn't have to be done for the yellow ball (refering to the one I used in the videos) as it is allready calibrated. If you wish to save the old parameters just rename the file threshold_matrix.xml!

Launch the programm with following command:

roslaunch tas hardware.launch
rosrun follow_ball navigation_goals_node

This will start the hardware of the car as well as the navigation_goals_node which will do the magic.

Connect the Wii-Controller and press the Nunchuck C-Button to allow the node to publish /servo.

With the yellow ball it works best in the cooridoors as in the lab there are many yellow things such as cables which might be detected as a ball. I think a red or green ball would generally work better in that environment.

in the file navigation_goals.cpp you can set the param use_autonomous to true/false

I tryed to implement that it checks the laser_scan ranges when autonomous is set to false but i couldn't test that anymore.

You can change parameters in the navigation_goals.cpp important params:

• DESIRED_R set it to the value depending on the size of the ball and the image (demo was 55 with a 640pixel width image, the ball had r ca. 1.5cm
• ANGLE_MAX depends on the viewing angle of the camera (set it a bit higher to be sure)
• MAX_SEC_TRY_TO_FOLLOW how long shall the car try to find the ball if it sees none (only important if autonomous is set to false)

If you wish to use the autonomous driving stack you should check if it makes any sense using the find_consistent_circle method of find_ball as well as the low processing rate set in navigation_goals.cpp.

Use this at your own risk as this functionality isn't tested!

1. Better algorithm to find a conistent ball.
2. If camera shall be used for other tasks as well determine angular speed with respect to the radius of the ball and the y-coordinate of the center of the ball. This way the camera can be put on top of the car.
3. Remove hardcoded parameters.
4. Improve use of laserscan so the car is capable of following the ball even if it is e.g near a wall.

If it doesnt find the threshold_matrix.xml, do catkin_make in a terminal, then start the navigation_goals_node or calibrate_params in the same terminal

Make sure you edit the servo-command topic such that it fits your robot if you wanna directly publish on the servo. Also you most likley have to change the callback if the wii-button is pressed. Everything is in the file navigation_goals.cpp.