Adapting to Perceived Safety within Human-Drone Interaction : Explored in proximity space

Sammanfattning: During recent years, drones’ presence in society has grown. They have many use cases, such as capturing video footage, delivering packages, protecting farmers’ land against wildlife, and fighting fires. Even though the amount of interactions differ, the drones somehow interact with humans in these use cases. This interaction should be perceived as safe for humans since feeling safe is such a basic need that humans have, and without it, they would probably not want to take part in the interaction. Therefore, this study was conducted to research how Control Barrier Functions (CBFs) could be used to improve perceived safety within Human-Drone Interaction (HDI). The study was conducted in a proximity space to control the interaction fully, both in terms of drone movement and how the participants were positioned. The study’s goal was to see if CBFs could be combined with Gaussian Processes (GPs) to personalize drone movement for the participants to feel safe while the drone completes its task. The GPs were used to learn what the humans thought was a safe drone trajectory, and the CBFs were used to (1) guarantee that the drone would never crash into the human and (2) allow the GPs to adjust how hard the drone was allowed to brake and how close it was allowed the human. The personalized drone movement was compared with two general drone movements, and in the end, the participants preferred their personalized trajectory over the two general ones. These results are interesting because it shows that drones could adapt to people in their surroundings to make them feel safer.