Cooperative behavior acquisition in multi robots environment by reinforcement learning based on action selection level

In a multi robots environment, the overlap of actions selected by each robot makes the acquisition of cooperation behaviors less efficient. So we propose an approach to determine the action selection priority level based on which the cooperative behaviors can be well controlled. First, we define eight levels for the action selection priority, which can be correspondingly mapped to eight subspaces of actions. Then by the local potential field method the action selection priority level for each robot is calculated and thus its action subspace is obtained. Third, reinforcement learning (RL) is employed to choose a proper action for each robot in its action subspace. Finally we have applied the proposed method to a soccer playing situation and the efficiency was verified by the results of both the computer simulation and real experiments