Imitation of Human Motion by Low Degree-of-Freedom Simulated Robots and Human Preference for Mappings Driven by Spinal, Arm, and Leg Activity

摘要

Robots cannot exactly replicate human motion, especially for low degree-of-freedom (DOF) robots, but perceptual imitation has been accomplished. Nevertheless, the multiple mappings between human and robot bodies continue to present questions around which aspect of human motion robots should preserve. In this vein, this paper presents a methodology for mapping human motion capture data to the motion of a low-DOF simulated robot; further, empirical experiments conducted on Amazon Mechanical Turk illuminate human preference across several such mappings. Users preferred motion capture driven robot motion over artificially generated robot motion, suggesting imitation was successfully accomplished by the proposed mappings. Moreover, one mapping, relating to the leaning of the spine, was preferred over arm- and leg-based mappings by a significant subgroup of respondents who were both loyal to the mapping-across multiple stimuli-and more engaged in the survey than other respondents. These results re-confirm the ability for simple robots to imitate human behavior and indicate that monitoring human spinal activity may be especially useful in this pursuit. Parallel work in psychology and human behavior analysis suggests that successful imitation of the motion of human counterparts is a necessary activity for robots to integrate in human-facing environments.