Analyzing people's complex task teaching process to a robot

Goal

Propose robot interaction design that makes teaching more accessible and efficient for a lay teacher

Research Questions

1. How do people teach complex tasks to a robot?
2. What challenges do they face while teaching such a task to the robot?

Task of Focus

Teach a robot how to bake a pizza in an oven in a simulated robot environment

Methods

Contextual Inquiry, Prototype Testing, Usability Study

Analysis Methods

Affinity Diagramming, Reflexive Thematic Analysis

Population

14 participants - diverse participants from the Ann Arbor Area recruited until data saturation

Select study results

• People developed complex teaching strategies
• People leveraged different types of knowledge available in the system
• People became effective at teaching over time
• Insufficient knowledge led to incomplete and incorrect mental models of the robot

  P2: “I think it’s stuck in some sort of weird loop with me wherein anything I do it just keeps asking for the next goal or sub task, but I feel my request, it’s fairly clear.”

  P10: “don’t know how many like arms the robot has. So I’m trying to open the oven with the pizza being held.”

• Incomplete mental models led to uncertainty and failures

  P8: “still trying to figure out whether the oven is preheated or not. I don’t see any actions in the list that give me that seem to give me an option to ask Rosie to check for me.”

  P6: “Man, I feel like I’m failing. I feel like there might be no answer to this.”

• However, people were motivated to continue teaching and found ways to recover from failures

Paper

Preeti Ramaraj. "Analysis of Situated Interactive Non-Expert Instruction of A Hierarchical Task to a Learning Robot." Doctoral Dissertation, University of Michigan, Ann Arbor. 2023.