Assessing the Perception of Human-Like Mechanical Impedance for Robotic Systems

Abstract

As physical interactions between robots and humans become more common, there is a growing need to design robots that are kinesthetically perceived as human-like. One approach to implement human-like mechanical impedance is to physically simulate models of the human neuromuscular system. However, the level of model complexity needed to achieve perception of human-like properties is unknown. The purpose of this study was to develop an objective assessment of a human subject's ability to discriminate kinesthetically between a model-defined impedance and that produced by human muscle. The assessment was based upon signal detection theory, by which the ability to discriminate between two classes of stimuli is analyzed statistically. With this assessment, we tested the hypothesis that a nonlinear muscle model is necessary to obtain perception of human-like muscle mechanical impedance. Fifteen subjects were presented with a mechanical impedance of either a simulated muscle model or electrically stimulated wrist muscle and were asked to decide if they were interacting with a “machine” or “human.” The impedances were randomized for a total of 30 presentations. The results showed that a robot that stimulates either linear viscoelastic properties or a nonlinear Hill muscle model could be distinguished from a human wrist muscle by almost all subjects. However, the subjects' ability to discriminate between the Hill model and human muscle was significantly less, which may have been due to larger overall impedance of the viscoelastic model. These results are important for the design of robots that emulates mechanical impedances that are perceived as human-like.