Explicit model identification and control of a micro aerial vehicle

Abstract

In this work, we provide a nonlinear mathematical model identification methodology of an autonomous micro quadrotor, and the design of its orientation and position controllers. In the model identification, we specifically focus on the brushed D.C. motor dynamics, which further breaks down into three different segments: voltage generation, motor dynamics, and force/torque generation. Test bench experiments and software simulation are conducted to identify the pa- rameters of the model derived from first principles physics model. Upon obtaining a good mathematical model of the micro quadrotor, model based orientation and position controllers are respectively implemented with linear quadratic regulator (LQR) and robust and perfect tracking (RPT) controller. The proposed control structure is designed and realized in a low cost micro quadrotor codenamed KayLion developed by the National University of Singapore. I. INTRODUCTION The recent progress in sensory technology, processing unit and integrated actuators has realized smaller unmanned aerial vehicle (UAV) platforms with higher intelligence lev- els. These kind of platforms, commonly known as micro aerial vehicles (MAVs) make both military and civilian tasks possible, such as indoor navigation, search and rescue, surveillance and reconnaissance, as well as potential danger detection. These tasks require a self-stable system with its own navigation and obstacle avoidance capabilities in different environments. However, based on a recent survey in (12), most commer- cial products of MAVs are not capable of carry powerful sensors and processors for autonomous flight. Due to the payload limitation, many of these small aircrafts do not have the endurance required for long-range missions. Most of the MAV platforms utilize ultra-light miniature sensors and Lithium-Polymer (Li-Po) batteries due to their light weight. As a result, relatively low-quality measurements and battery endurance limit the performance of the MAV in navigation and localization. A recent example sees a palm-sized gliding MAV developed by Harvard University weighing only 2 g and 10 cm in wing-span. It is capable of autonomous flight target sensing and obstacle avoidance using an optical flow sensor. Nevertheless, without any propulsion system, the platform is only able to glide with an initial thrust (16). In MAV control and autonomous flight, a single cam- era as the sole position estimation sensor is preferred by