The basic skills of our agent are acquired by recording the human actions, segmenting the interesting parts, grounding the observed actions by the primitives of our agent, and adjusting the parameters in rehearsals. Besides, extra efforts are needed to make the acquired basic skills adapted to the different situations.
A representation of high-level strategies are developed in our agent. Based on it, strategies from one human is imitated. In addition, the optimal strategy in the current game, which yields the best performance on game scores, is given by the fast adaptation algorithms.
Our research is useful in the applications of dynamic competition games, which involve real robots, such as different leagues in RoboCup and table soccer robot as well as computer games such as Age of Empire and Tetris.