Modeling Trust to Improve Human-Robot Interaction.

摘要

Throughout the history of automation, there have been numerous accidents attributed to the inappropriate use of automated systems. Over-reliance and under-reliance on automation are well documented problems in fields where automation has been employed for a long time, such as factories and aviation. Research has shown that one of the key factors that influence an operator's reliance on automated systems is his or her trust of the system. Several factors, including risk, workload, and task difficulty have been found to influence an operator's trust of an automated system. With a model of trust based upon these factors, it is possible to design automated systems that foster well-calibrated trust and thereby prevent the misuse of automation.;Over the past decade, robot systems have become more commonplace and increasingly autonomous. With the increased use of robot systems in multiple application domains, models of trust and operator behavior for human-robot interaction (HRI) must be created now in order to avoid some of the problems encountered by other automation domains in the past. Since traditional automation domains and HRI are significantly different, we reexamine trust and control allocation (operator's usage of autonomous behaviors) as it relates to robots with autonomous capabilities in order to discover the relevant factors in HRI.;This dissertation examines existing work in traditional automation that is relevant to HRI and, based on that information, builds an experimental methodology to closely mimic real world remote robot teleoperation tasks. We also present results from multiple experiments examining the relationship between the different factors being investigated with respect to trust and control allocation. Based on these results, a model for human interaction with remote robots for teleoperation (HARRT) is proposed and design guidelines to help improve overall performance are presented based on the model.