Learning by Imitation for Playing Table Soccer in Dynamic, Noisy, and Unpredictable Environments

01, 12, 2005 – 30, 11, 2006

Table Soccer is a popular game usually seen in arcades and bars. Developed at the university of Freiburg, the table soccer robot controls one side of the table soccer game, playing with human players on the other side. The robot can challenge even advanced human players because it has faster reactions than most human players.

The proposed project is particularly interesting, mainly for the three reasons given as follows. Firstly, being normally used with humanoid robots, learning by imitation is studied with the robot which has less degree of freedom, faster speed, and higher accuracy than a humanoid robot. Secondly, the ball moving in a continuous space is taken into consideration in the imitation, while most applications using imitation only consider the movements of actuators. Finally, as being improved by learning by imitation, the robot would be more intelligent and amazing in the table soccer games.

Several steps are needed to realize the proposed project. Firstly, hardware improvements enable the robot to observe the movements of the human players and increase the accuracy of the ball recognition. Then, learning by imitation could be implemented by the segmentation and interpretation of the data, and action replication. The adaptation of the ball position could be considered in the data representation. Finally, to choose a proper action in the games, an action selection mechanism is implemented.

Supported by Karl-Steinbuch-Stipendium

Home | Publication