Do humans need learning to read humanoid lifting actions?

Abstract

Humans can infer, just from the observation of others' actions, several information on the actor's intents and on the properties of the manipulated objects. This intuitive understanding is very efficient and allows two collaborating partners to be prepared to handle the common tools, as they can estimate the weight of the object the other agent is passing to them even before the hand-over is concluded. Transferring this kind of mutual understanding to human - robot interactions would be particularly beneficial, as it would improve the fluidity of any collaborative task. The question that we address in this study is therefore under which conditions humans can estimate the weight lifted by a humanoid robot and whether the acquisition of this skill requires an extensive learning by the human subject. Moreover, we assess whether reading humanoid lifting actions implies the involvement of the observer's motor system, as it happens for weight judgment from the observation of human actions. Our results indicate that with a proper design of the humanoid lifting motions, human subjects are able to estimate the weight lifted by the humanoid robot with a similar accuracy as that exhibited during human observation. Furthermore, such ability is intuitive and does not require learning or training. Lastly, weight judgment seems to be dependent on the involvement of the observer's motor system both during human and humanoid observation. These findings suggest that the neural mechanisms at the basis of human interaction can be extended to human-humanoid interaction, allowing for intuitive and proficient collaboration between humanoid robots and untrained human partners.