Nonlinear Vision-Based Observer for Visual Servo Control of an Aerial Robot in Global Positioning System Denied Environments

This paper presents a nonlinear vision-based observer to estimate 3D translational position and velocity of a quadrotor aerial robot for closed-loop, position-based, visual-servo control in global positioning system (GPS)-denied environments. The method allows for motion control in areas where GPS signals are weak or absent, for example, inside of a building. Herein, the robot uses a low-cost on-board camera to observe at least two feature points fixed in the world frame to self-localize for feedback control, without constraints on the altitude of the robot. The nonlinear observer described takes advantage of the geometry of the perspective projection and is designed to update the translational position and velocity in real-time by exploiting visual information and information from an inertial measurement unit. One key advantage of the algorithm is it does not require constraints or assumptions on the altitude and initial estimation errors. Two new controllers based on the backstepping technique that take advantage of the estimator's output are described and implemented for trajectory tracking. The Lyapunov method is used to show asymptotic stability of the closed-loop system. Simulation and experimental results from an indoor environment where GPS localization is not available are presented to demonstrate feasibility and validate the performance of the observer and control system for hovering and tracking a circular trajectory defined in the world frame.