GENERALIZATION OF HUMAN TRUST IN ROBOT CAPABILITIES ACROSS TASKS

Abstract

There has been extensive study in human-robot interaction. One of the most important topics in HRI domain is understanding human trust in robotics. We found that the existing study either focused on modeling trust as a single measurement for different tasks, or limiting it to one set of tasks. We extend the previous study by investigating how trust generalizes across tasks of different categories and difficulties with observations of robot performance. We conducted human-subject experiments in two different setups - household tasks with Fetch Research Robot for picking and placing objects and indoor navigation, and autonomous driving tasks with VR simulation system. The experimental results show that human trust generalization across tasks significantly depends on the perceived task similarity, perceived task difficulty and robot performance. In addition, we also studied the direction and scale factor of trust when trust gained from certain task(s) generalizes to other tasks.