부산대학교 도서관

A theoretical control solution to improve the navigation performance of an aerial vehicle, focusing on a quadcopter vehicle, is presented in this article. The main control purpose is to endow the aerial vehicle with predictive and reactive properties (evasive maneuvers) to avoid collisions with dynamic objects (or other aerial agents) that may be in direct collision with it. The proposed control architecture is formed by three modules: a predictor algorithm, an upper level command generator, and a lower level drone controller. The trajectory predictor algorithm gives a horizon of possible future positions where an object could collide with the aerial vehicle. These predictions are sent to the second module, which, based on an artificial particle field, generates velocity commands to avoid possible collisions. These commands are then sent to the drone controller, which was designed using the Lyapunov formalism to guarantee an agile and stable response to the aggressive nature of the problem. The overall architecture was validated and tested in real time with different scenarios and using a cyber-physical twin framework, showing good capabilities to anticipate (predictive properties) and avoid (reactive properties) imminent collisions.