Predictive Minimum Variance Control: a Case Study on Wireless Altitude Hold Autopilot

Abstract

Unmanned aerial vehicles, or UAVs, used in collaborative missions, are commonly supervised in flight by a master guidance, navigation and control system that must verify that every flight vehicle is performing the mission as prescribed. Within this scenario, the control system may be separated from the controlled UAVs by great distances, leading to data loss, increased transport time delays and reduced control stability margins. In this work this scenario is reproduced experimentally using a single UAV quadcopter and a remote master control unit connected by a wireless network. By indoor laboratory flight measurements a considerable time delay was verified and embedded in the UAV thrust to altitude estimated model. Then, a long-range prediction horizon control solution, based on the Generalized Minimum Variance Control in the State-Space, was investigated and its results compared to a benchmark controller. The predictive controller outperformed the benchmark one with respect to reference tracking dispersion minimization and control signal chattering reduction in the task of controlling the altitude of a quadcopter over a wireless network in real flight experiments. This work also discusses further insights on predictive minimum variance control from the perspective of the so-called general Bolza optimal control problem in order to remark the differences from most common model predictive control techniques.