A dynamic neural field approach to natural and efficient human-robot collaboration

Abstract

A major challenge in modern robotics is the design of autonomous robots that are able to cooperate with people in their daily tasks in a human-like way. We address the challenge of natural human-robot interactions by using the theoretical framework of dynamic neural fields (DNFs) to develop processing architectures that are based on neuro-cognitive mechanisms supporting human joint action. By explaining the emergence of self-stabilized activity in neuronal populations, dynamic field theory provides a systematic way to endow a robot with crucial cognitive functions such as working memory, prediction and decision making . The DNF architecture for joint action is organized as a large scale network of reciprocally connected neuronal populations that encode in their firing patterns specific motor behaviors, action goals, contextual cues and shared task knowledge. Ultimately, it implements a context-dependent mapping from observed actions of the human onto adequate complementary behaviors that takes into account the inferred goal of the co-actor. We present results of flexible and fluent human-robot cooperation in a task in which the team has to assemble a toy object from its components.