A Method to Determine Human-Likeness in Social Motions of Anthropomorphic Robots

Abstract

The hand waving motion of a human was considered as a representative social motion because a human can use such motions for social interaction and communication. Twenty healthy human subjects were recruited to participate in a study, where each subject separately showed the hand waving motion at three different conditions: (i) both hands waved, (ii) only left hand waved, and (iii) only right hand waved. The hand waving motion was captured by a motion capture system for each subject in each condition separately. Then, the kinematics (absolute linear displacements, velocities and accelerations) along different axes at three different joints such as wrist, elbow and shoulder were analyzed. Then, the hand waving motion was generated in two different anthropomorphic robotic platforms where the robots were enabled to show their hand waving motions at the same three conditions. The kinematic features for hand waving of the robots along different axes at three different joints were captured using the motion capture system and analyzed in the same way as it was done for the humans. Then, a dynamic similarity metric called the Froude number was proposed and used to determine human-likeness in the form of dynamic equivalence between human and robot motions. Human-likeness between human and robot motions were also assessed through a human subject study to crosscheck the results obtained through the use of the Froude number. An agreement was found in the results obtained through two different methods (Froude number, subjective study). The proposed approach can help determine human-likeness of motions generated by anthropomorphic robots that can bring balance in human-like appearance and human-like motions or actions of anthropomorphic robots and virtual characters, which can enhance their chance of being accepted by their human counterparts for coexistence and collaboration.