Determining Human Performance Given Robot-Provided Explanations Through Task-Based Interaction
Open Access
- Author:
- Hannah, Sydney
- Area of Honors:
- Mechanical Engineering
- Degree:
- Bachelor of Science
- Document Type:
- Thesis
- Thesis Supervisors:
- Alan Richard Wagner, Thesis Supervisor
Bo Cheng, Thesis Honors Advisor
Katie Fitzsimons, Faculty Reader - Keywords:
- human-robot interaction
task-based interaction
simulation
explanation
robot
human performance - Abstract:
- As the applications and abilities of robots in society continue to progress, a thorough understanding of a robot’s interaction with humans is critical. In practice, human-robot teams will need to communicate effectively in order to accomplish tasks and larger goals. Task-based goals that involve decision-making from one entity often require explanations to justify actions or ensure understanding among team members. This study investigates human performance based on the explanations provided by a robot that are intended to describe how the robot sorted a series of blocks and based on participant-created explanations. In this study, explanation is defined as communicating with the intention of describing information, which is in this case how a series of blocks was sorted into a pattern. Participants were guided by a robot in a simulation in which they were asked to identify the correct pattern based off the robot-provided explanation and rate quality of the robot’s explanation in order to analyze participant performance. Participants were also asked to create their own pattern of blocks and explain it to the robot to measure matching of communication style. Our findings indicate that participants were able to consistently distinguish between good, medium, and poor explanations, and rated the quality of explanations in the expected order. Surprisingly, we find that participants correctly identified patterns explained by the robot when the type of explanation provided was poor, and misidentify the pattern the most often when the type of explanation provided was of medium quality, which can be explained in part by the number of factors by which the patterns were sorted. Such findings cannot explicitly determine the relationship between human performance and quality of explanation, but still provide valuable insight into human performance in task-based interactions. Additionally, participant-generated patterns and explanations offer insight on the convergence of linguistic and communication style and present ample opportunities for future work in human-robot communication on a psychological level. The findings from this experiment present important considerations for future research and utilization of robots in applied settings in which comprehensible explanation is pertinent.