Vision-based Model Predictive Control for Unmanned Aerial Vehicles Automatic Trajectory Generation and Tracking

This paper proposes a single-feature vision-based control system for unmanned aerial vehicles with hovering capability. The purpose of the control system is to perform a desired maneuver while keeping a specified visual feature in a certain position of the image plane. Since the feature tracking problem constitutes a non-linear static constraint on the system dynamics, a theoretical analysis of the equilibrium points of the system is presented, and a methodology for exploiting this result for trajectory generation is proposed. Knowledge of the loci of equilibrium points of the system, which depends on which variables are selected as dependent and independent, allows to generate a large variety of vehicle trajectories without actually using the vehicle position as variable to be regulated. Model Predictive Control is used to have the vehicle actually perform the desired maneuver while enforcing excursion constraints on relevant state and control variables. Examples are proposed for hovering relatively to a target (the visual feature), orbiting at constant altitude or spiraling around a target, and landing on a target with a desired flight path angle; all these trajectories are obtained by manipulating a minimal set of state and output variables, and without employing any guidance, trajectory tracking or path following algorithm.