My thesis extended an existing study on human-robot shared control. A person shared control of a flying robot in a 2D simulation and could take over whenever they didn't like how it was steering. We treated intervention as a signal of trust: more takeovers meant the robot was steering poorly, so the controller tried to fly in ways that reduced them.
Stack: C++, OpenCV, Python
What I added
A voice. When the controller's confidence dropped below a threshold, it played a human-voiced audio cue saying it was unsure. I measured how those cues changed intervention, comparing agents with and without the warnings, and with male and female voices. Trust was measured two ways: inferred from interventions, and self-reported by participants.
What we found
Verbal feedback from a robot changes how much people trust it. People also tended to trust agents whose voice matched their own gender.
Published at CRV 2019
The paper, Investigating Trust Factors in Human-Robot Shared Control: Implicit Gender Bias Around Robot Voice, was accepted to the Conference on Computer and Robot Vision, and I was flown out to Queen's University to present it. According to the organizers, it was the first study of gender bias in the conference's history.