Privacy-preserving feature extraction for HRI and effect on self-disclosure

Most social robots are equipped with sensors that allow them to gauge what is happening in their environment [1]. Commonly used sensors are related to the three most commonly used modalities in human interaction: vision, audition and touch [1]. The presence of such sensors in a person’s space raises privacy concerns and could lead to distrust of the robot [2]. Additionaly, several studies in the field of computer-mediated communication showed the importance of visual anonymity in human self disclosure [3]. The aim of this work is twofolds. First, building upon existing work [4], we aim to develop methods to extract human features (human pose, face detection, gender recognition, age recognition, emotions recognition, etc) out of privacy preserving images, i.e. images blurred using a physical filter. Second, through a user study, we aim to explore the effect of robot transparency, i.e. the act of sharing to a human interlocutor what a robot perceives, as well as the effect of sharing the use of privacy-preserving sensing, on human self-disclosure.

Concretely, in this work, you will: (1) review the existing literature on privacy preserving feature extraction for human-robot interaction; (2) integrate already existing privacy- preserving human pose estimation in a ROS package; (3) improve and extend the aforementioned work to extract other human features; (4) build a dataset of images (with and without privacy-preserving filtering) to compare the detection performance of the extracted human features; (5) design a simple experiment with human participants to explore the effect of robot transparency and privacy-preserving sensing on human self-disclosure; (6) summarise the work carried out in a scientific report. This work requires you to have experience with the Python programming language. Familiarity with ROS, deep learning techniques for computer vision and code versioning (Git/GitLab) can be advantageous.

This work will give you the opportunity to (1) gain experience in Python and ROS
programming; (2) learn how to conduct user studies; (3) write and potentially publish a
scientific article to a conference in social robotics.

Starting date: As soon as possible

Supervisor(s): Romain Maure (romain.maure∂kit.edu), Barbara Bruno (barbara.bruno∂kit.edu)

References:
[1] Bartneck, C., Belpaeme, T., Eyssel, F., Kanda, T., Keijsers, M., & Sabanovic, S. (2020). Human-Robot Interaction – An Introduction. Cambridge: Cambridge University Press

[2] Caine, Kelly, Selma Šabanovic, and Mary Carter. "The effect of monitoring by cameras and robots on the privacy enhancing behaviors of older adults." Proceedings of the seventh annual ACM/IEEE international conference on Human-Robot Interaction. 2012.

[3] Joinson, Adam N. "Self‐disclosure in computer‐mediated communication: The role of self‐awareness and visual anonymity." European journal of social psychology 31.2 (2001): 177- 192.

[4] Xia, Youya, et al. "Privacy-preserving pose estimation for human-robot interaction." arXiv
preprint arXiv:2011.07387 (2020).