IDrone

Swarms of Unmanned Aerial Vehicles (drones) could provide great benefit when used for disaster response and indoor search and rescue scenarios. In these harsh environments where GPS availability cannot be ensured, prior work often relies on cameras for control and localization. This creates the challenge of identifying each drone, i.e., finding the association between each physical ID (such as their radio address) and their visual ID (such as an object tracker output). To address this problem, prior work relies on visual cues such as LEDs or colored markers to provide unique information for identification. However, these methods often increase deployment difficulty, are sensitive to environmental changes, not robust to distance and might require hardware changes. In this paper, we present IDrone, a robust physical drone identification system through motion matching and actuation feedback. The intuition is to (1) identify each drone by matching the motion detected through their inertial sensors and from an external camera, and (2) control the drones so they move in unique patterns that allow for fast identification, while minimizing the risk of collision involved in controlling drones with uncertain identification. To validate our approach, we conduct both simulation and real experimentation with autonomous drones for the simplified case of a stationary Spotter (powerful drone equipped with the camera). Overall, our initial results show that our approach offers a great tradeoff between fast identification and small collision probability. In particular, IDrone achieves faster identification time than safety-based baseline actuation (one-at-a-time), and significantly higher survival rate compared to fast, non-safety-based baseline actuation (random motion).