Relative localization with computer vision and UWB range for flying robot formation control

Relative localization is a core problem for swarm robotics since each swarm node must determine where neighboring robots are located to accomplish cooperative tasks such as formation control. In this paper we present a system that performs relative localization between a stationary marker and a flying robot in real-time. Relative distances are obtained using ultra-wide band radio, while a low-cost webcam provide angle measurements. To achieve the latter, we employed the Camshift algorithm and a Kalman filter. We tested our system outdoors during daylight using centimeter-accuracy GPS measurements as ground truth. Three data sets have been collected from a series of experiments and it shows that errors in estimated relative positions are between +/-0.190 m on the x-East axis and +/-0.291 m on the z-North axis at 95% confidence level.