Robust Image-based Visual Servoing of an Aerial Robot Using Self-organizing Neural Networks

The problem of robust image-based visual servoing of an aerial robot for tracking a moving vehicle is addressed in this brief. First, a camera perspective projection model is used with a coordinate frame attached to the quadrotor with a fixed relative pose with respect to it. Next, several world points are projected onto the image plane so that features may be extracted from the projected point coordinates of the goal object. The chosen features are functions of the zeroth, first, and second moments of the acquired binary image. These features provide signals associated with the position and yaw angle of the aerial robot in Cartesian space. The visual feature error vector generates the desired thrust vector that stabilizes the position and yaw angle error of the platform. However, significant uncertainty is imposed on the system leading to a negative impact on the tracking performance of the control system. This paper contributes to the field by introducing a self-organizing neural network (SONN) that can circumvent the challenge stemming not only from the changes in the velocity of the moving vehicle but also payload variations. Additionally, extensive ROS Gazebo simulations and real-world experiments are conducted to assess the effectiveness of this method, in contrast to many existing studies that rely solely on MATLAB-based simulations. Eventually, the stability of the closed-loop system is proven through Lyapunov theory with the uncertainty terms taken into account.