Action Selection in Robots Based on Learning Fuzzy Cognitive Map

One of the main issues in developing automatic response systems especially autonomous robots is selecting the best action among all possible actions. Fuzzy cognitive maps (FCMs) aim to mimic the reasoning process of the human. FCMs are able to capture and imitate human behavior by describing, developing and representing models. FCMs are also popular for their simplicity and transparency while being successful in a variety of applications. We developed a novel model that could be used for action selection in robots. This model is constructed on a learning FCM which is relied on improved nonlinear Hebbian algorithm. We tested our model through a series of practical experiments on the latest version of soccer server simulation 3D environment. Our tests involved carefully defined factors to measure the team performance. Our results showed a significant improvement in overall performance.