NAO learns to push the box controlled by a Learning Classifier System

Описание: This is the first version of NAO navigation by pushing box. The destination is 10 cm ahead, covered by the box. So the NAO must push the box to achieve the goal. It has 4 discrete actions to choose : move left 10cm, move ahead 10cm, move ahead 10cm while pushing, and do nothing. Only the pushing one is effective to the move, although the box is movable by simply kicking it while walking ahead, which is not encouraged.
Exploration/Exploitation framework in XCS is used, learning only happens in explore trails. Finally, the NAO learns the rule(move ahead 10cm while pushing) which gives the highest prediction of reward and has highest fitness.