Mirror neurons are premotor neurons that are considered to play a role in goal-directed actions, action understanding and even social cognition. As one of the promising research areas in psychology, cognitive neuroscience and cognitive physiology, understanding mirror neurons in a social cognition context, whether with neural or computational models, is still an open issue [5]. In this paper, we mainly focus on the action understanding aspect of mirror neurons, which can be regarded as a fundamental function of social cooperation and social cognition. Our proposed initial architecture is to learn a simulation of the walking pattern of a humanoid robot and to predict where the robot is heading on the basis of its previous walking trajectory.