Enabling Common Ground in Human-Robot Interaction

Goal

Modify robot interaction mechanisms to improve people's teaching experience

Research Questions

1. Can lay teachers teach a robotic agent to do simple tasks via verbal instructions?
2. What knowledge do they need from the robot to teach easily and effectively?

Task of Focus

Teach robot to solve a Tower-of-Hanoi puzzle via verbal instructions and demonstrations

Methods

Wizard-of-Oz testing, Prototype/Formative Usability Testing, Surveys

Analysis Methods

Qualitative Analysis, Cognitive Task Analysis

Population

10 students from the University of Michigan

Select study results

Analysis of survey responses

• Participants got the hang of teaching the robot over time.

  P1: "I liked that talking to her (the robot) became simple after a few interactions"

• Participants appreciated the robot's feedback after every instruction.

  P8: "It was nice to be able to get feedback whenever I tried giving the agent an instruction (i.e., whether it understood the instruction or not)."

• However, some participants found the feedback insufficient when the robot failed to understand or execute their instructions

  P4: "Responses such as "I don't understand first action mentioned" or something to let me know what part of the statement Rosie (the robot) doesn't understand."

Analysis of individual teaching interaction transcripts

Through a task analysis of the individual teaching interaction, we found that participants ran into multiple types of failures. The primary failure was when the robot was unable to provide sufficient information about how it perceives relationships between objects in the environment. This revealed a need for a visual transparency mechanism that allows a teacher to access the robot's perception of the world.