As teleoperated robot technology improves, remotely-operated telepresence robots will certainly become more prevalent in homes and businesses, allowing guests and business partners to visit without being physically present. That said, privacy remains a subject of concern. An Internet-connected telepresence robot has the ability to, unbeknownst to the robot’s owner, spy on its local area. Whether by the remote operator or by a third party with access to the video data, telepresence robots represent a unique vulnerability to our privacy.
Existing solutions to this problem rely on various algorithms to detect objects, identify them as private or not, and apply protections accordingly. While this approach has merit, it requires a high-quality camera and sufficient computing resources to run the algorithms in real-time. Further, if the algorithms fail to identify or properly protect an otherwise-private object, even for a single frame, a person with access to a recording of the video would still be able to breach privacy.
As a possible solution for these concerns, we decided to apply protections to the whole video feed. In addition to the robustness of this approach, rendering the latter concern moot, we found that these image manipulations could still run quickly on a rather weak computer. Three such video filters were created, making use of the Kinect present on our remotely-operated Turtlebots.
We conducted a user study to examine the effects of applying these whole-image filters on robot usability and privacy protection. We also looked at the effects of cognitive load and robot operation, as it is possible that the operator is simply too distracted to breach privacy in the first place, though this would obviously have no effect on a third-party with a video recording.
We found that applying such filters protected privacy without significantly affecting the robot’s usability and that the video feed from the depth camera was the most effective privacy protector. We also found that the cognitive load of driving the robot has a slight privacy-protecting effect.
Jeffrey Klow, ’17
Jordan Proby, ’19