Experiment protocol
While existing research has examined many of the same factors as our project, our research brings an aspect of novelty by specifically focusing on mirrored eye gaze while holding other factors constant. Many studies have researched eye gaze in conjunction with other nonverbal cues, such as neck movements, head tracking, and posture. In addition, most of these studies use trust as the dependent variable, but we targeted comfort level as our subjective metric.
- Participants complete a pre-study questionnaire with demographic information, and give consent.
- Participants calibrate eye-tracking glasses by introducing themselves to Talent, the robot.
- Talent begins part 1 of the interview with participants randomly assigned to start either the control condition (no eye gaze mirroring or the experimental condition (eye gaze mirroring).
- Participant completes the post-study questionnaire for part 1.
- Talent begins part 2 of the interview with the alternate condition (experimental vs. control).
- Participant completes the post-study questionnaire for part 2.
Task Script
The task script, my primary responsibility, was designed to facilitate a conversation between “Talent” and a “job seeker,” i.e. the participant. Using VoiceFlow, a conversational assistant design platform, we developed several conversational paths that were later executed during the study as the voice of “Talent,” using a Wizard of Oz technique. The task script was broken into three segments, all consisting of common interview questions.
Introduction to Talent
The first segment served as a general introduction and allowed for a baseline capture of eye gaze data from the participant to calibrate the eye-tracking glasses.
Control/Experimental
Segments two and three, designed to follow the natural cadence of an interview, prompted users to answer questions. These served as the basis for the control and experimental groups.
Natural utterances
Throughout“Talent” would respond to the participants' with utterances such as “That’s a great school!” and “Thank you for sharing.” to more closely mimic the natural exchanges of conversation.
Conversation paths
Using VoiceFlow, we were able to build multiple conversation paths in advance and select the path unique to the participants desired job title, while remaining consistent in the questions asked.
Robotics system development
Built by Gerry D'Ascoli and Michelle Lu, the experiment software is broken into three separate python scripts. The first interfaces with the Pupil Labs eye tracking Pupil Core glasses, the second script using Pygame (after pyglet proved unable to fit our automated update technical need) and the third script takes the models built up during each trial based on the subject’s gaze and loads them into a csv file for analysis.
Pupil Core Interface
The Pupil Core interface connects to pupil capture, subscribes to and reads the gaze messages published by the glasses, and builds the model based on the user’s gaze pattern from which we pull our objective metrics.
- Connected to Pupil Capture
- Read pupil gaze messages
- Build gaze model for objective metrics
Robot Animation
The robot animation script was built in Pygame. It is essentially two images overlayed, one of the robot and one of the robot’s pupils.
- Built in Pygame
- Two images overlayed
- Eyes moved based on input polar coordinates (r,θ)
After exploring different robot options in the AI Maker Space at Carnegie Mellon University, we ultimately decided to display our robot using computer graphics rather than a physical robot due to the limited access we had to robots whose eyes could be manually programmed.
Survey design
Designed by Sophia Timko, participants were asked to complete a survey after each trial was based on the 5-point Likert scale. Many of the questions were derived from RoSAS about comfort. To manage bias, the survey wording frames half the survey question positively and half negatively. To allow subjects to elaborate more, we asked 3 open-ended questions. and give opportunity to bring up factors that the survey didn't consider.
