Effective Task Training Strategies for Instructional Robots

Abstract

From teaching in labs to training for assembly, a key role that robots are expected to play is to instruct their users in completing physical tasks. While task instruction requires a wide range of capabilities, such as effective use of verbal and nonverbal language, a fundamental requirement for an instructional robot is to provide its students with task instructions in a way that maximizes their understanding of and performance in the task. In this paper, we present an autonomous instructional robot system and investigate how different instructional strategies affect user performance and experience. We collected data on human instructor-trainee interactions in a pipe-assembly task. Our analysis identified two key instructional strategies: (1) grouping instructions together and (2) summarizing the outcome of subsequent instructions. We implemented these strategies into a humanlike robot that autonomously instructed its users in the same pipe-assembly task. To achieve autonomous instruction, we also developed a repair mechanism that enabled the robot to correct mistakes and misunderstandings. An evaluation of the instructional strategies in a human-robot interaction study showed that employing the grouping strategy resulted in faster task completion and increased rapport with the robot, although it also increased the number of task breakdowns. Our model of instructional strategies and study findings offer strong implications for the design of instructional robots.