Tracking and Navigation for Soccer Robots
During my master's study at the University of Bonn, I was actively interested in robot learning after working on several course projects related to robotics. My master thesis aims to build a more robust visual tracking system that allows the humanoid soccer robot to achieve better performance in motion planning in RoboCup competition. I was a member of RoboCup team NimbRo TeenSize at Univ. of Bonn. My thesis work helped our team won the 4th place in the Humanoid League competition and the 1st place RoboCup Design Award.
As for a soccer robot in the RoboCup competition, It is crucial to detect field features efficiently for matching to get an accurate estimate of its poses. The previous approaches only realized three degrees of freedom (3-DoF) of robot pose using visual perception and retrieved another 3-DoF pose from IMU and Kinematic model. Since the IMU data is not always reliable, it may reduce the accuracy of the resulting estimation. To address these challenges, I proposed a new kernel-based line detection method for extracting field features from the input images.
Besides, I also developed a model-based visual tracking system that can track 6-DoF camera pose on the soccer field. The experiment shows that the line detection method is less sensitive to light condition changes than the color thresholding-based approaches and the system is less vulnerable than traditional approaches to the error from IMU data.
The soccoer field in RoboCup 2015 at Heifei, China
We won design award!
NimbRo Team in RoboCup 2015