Data-Driven Motion Mappings Improve Transparency in Teleoperation

Abstract

A teleoperation system with high transparency enables the operator to focus on completing the task at hand instead of on controlling the robot. We previously proposed that modifying the mapping from human movement to desired robot movement might improve the transparency of teleoperators in ways similar to adding sensory feedback. Specifically, we created non-Cartesian motion mappings that correct for systematic reaching errors made by humans, so that the robot motion resembles the operator's intent rather than his or her produced movement. This article presents a study that compares subjects' performance on a virtual teleoperated targeting task under three different motion mappings: a Cartesian-scaling motion mapping that is typically implemented in teleoperators, a corrective variable-similarity motion mapping that is fit to aggregate data from subjects in a previous study, and a corrective variable-similarity motion mapping that is fit to calibration data collected from each subject. Twelve participants reached toward 120 targets under each of the three motion mappings with balanced random presentation order and a washout task between conditions. Subjects were able to complete the targeting task with higher accuracy in the initial direction of robot motion, at higher speeds, and with more natural and efficient reaching movements under the variable-similarity motion mappings. Subjects also overwhelmingly preferred the variable-similarity motion mappings. These results indicate that subjects experienced a higher level of transparency when using the virtual teleoperator with the variable-similarity motion mappings than with the standard Cartesian mapping. Therefore, mappings that correct for systematic errors in human motion, such as the variable-similarity motion mappings tested here, should be considered in teleoperator design.