Robotic Model of the Mirror Neuron System: A Revival

Action understanding is a vital cognitive capacity for robots interacting with humans. The role of mirror neurons in action understanding has been widely discussed and modeled in the field of cognitive robotics. A mirror neuron system (MNS)-based neural machinery allows the robot to link the high-level representation of the observed action with a high-level representation of the same or similar action within its own motor repertoire and thus facilitate the understanding of the observed scene. We present a novel version of a multi-layer connectionist model of MNS for a cognitive robot that connects visual and motor modalities in a hierarchical fashion. It is based on an existing model for a humanoid iCub robot that learns to perform and understand grasping actions. We adapted our MNS model for the humanoid NICO robot. Our preliminary results indicate that our model is able to form mutual representation for perception and action and show us a promising perspective of developing action understanding and imitation in the future.