Towards a robotic model of the mirror neuron system

Abstract

Action understanding undoubtedly involves visual representations. However, linking the observed action with the respective motor category might facilitate processing and provide us with the mechanism to “step into the shoes” of the observed agent. Such principle might be very useful also for a cognitive robot allowing it to link the observed action with its own motor repertoire in order to understand the observed scene. A recent account on action understanding based on computational modeling methodology suggests that it depends on mutual interaction between visual and motor areas. We present a multi-layer connectionist model of action understanding circuitry and mirror neurons, emphasizing the bidirectional activation flow between visual and motor areas. To accomplish the mapping between two high-level modal representations we developed a bidirectional activation-based learning algorithm inspired by a supervised, biologically plausible GeneRec algorithm. We implemented our model in a simulated iCub robot that learns a grasping task. Within two experiments we show the function of the two topmost layers of our model. We also discuss further steps to be done to extend the functionality of our model.