IDENTIFICATION AND CONTROL OF A COMMERCIAL QUADROTOR HELICOPTER Cristian Souza, 1 Guilherme V. Raffo, 1 Douglas W. Bertol, 1 Eugenio B. Castelan 1 1 Departamento de Automação e Sistemas, Universidade Federal de Santa Catarina, CEP 88040-900, Florianópolis, SC, Brasil, {cristian, raffo, dwbertol, eugenio@das.ufsc.br} Abstract: This paper presents an identified model of a com- mercial quadrotor helicopter and a control design to perform tracking on the XY frame. The system is identified as a Hammerstein-Wiener model and the controllers are derived through the root locus method. Simulation results are carried out to validate the identified model and to corroborate the pro- posed controllers. Keywords: Quadrotor helicopter, Hammerstein-Wiener model, root locus method. 1 I NTRODUCTION The development of autonomous UAV’s (Unmanned Aerial Vehicles) has been gaining space on mobile robotics field in the last few decades due to both military and civil applica- tions. From these applications, tasks as search and rescue, remote inspection, mapping and surveillance can be high- lighted. In the flight control area, the most studied systems have been airplanes and standard helicopters (i.e., helicopters with main and tail rotors). However, in the last years, UAV’s in the quadrotor configuration have been featured in a lot of works [1–6]. Quadrotors are UAV’s that make use of four propellers to generate motion. Since this vehicle is based on VTOL (Vertical Take-off and Landing) concept, quadrotors have evident advantages over other aircrafts once they can take-off and land in limited areas, fly at low speeds and easily hover over targets. Moreover, they have high maneuverability. These features make possible to perform tasks that otherwise would be of great complexity, dangerous or even impossible. In addition, the construction of this kind of vehicle is simple and due to the fact that the propellers are fixed (absence of cyclic and collective commands) the maintenance becomes a trivial task. Despite all the advantages mentioned above, it is important to emphasize some drawbacks as the high energy consumption and the difficulty in controlling the quadrotor, since this kind of helicopter has highly nonlinear dynamics, strongly coupling, underactuation and open-loop instability. To control a nonlinear system, it is important to have knowledge about its behavior to compute a good mathemati- cal model of the real system dynamic. To obtain a good rep- resentation, it can be used a phenomenological model or an identification procedure to estimate the parameters of a pre- defined model. In [7] the parameters of a helicopter model are identified to achieve a LTI (Linear Time-Invariant) model operating at hover by using a PEM (Prediction Error Method) MIMO (Multi-Input-Multi-Output) estimation algorithm. In [8] the dynamics of a quadrotor is mathematically described based on Newton’s law. The inertia matrix, the aerodynamic friction coefficients and the translational drag coefficients are estimated by using pendulum devices. The lift constant and the drag coefficient of rotation are estimated experimentally, where the angular speed and the force generated by the pro- pellers are measured. The rotor parameters are approximated using quadratic optimization method based on the motor input and rotor angular speed measurements. Many efforts have been made to control quadrotor heli- copters and several strategies have been developed to tackle the tracking problem for this type of system. In [9] a neu- ral network was trained to make a four rotor helicopter capa- ble of achieving vertical take-off and landing and to sustain a specified attitude. First the training data is collected while the helicopter is manually piloted. In a second stage, a neural network using the collected data is trained. In the last phase, the parameters of the trained neural network are transfered to a similarly configured neural network in the helicopter. In [10] a quadrotor is modeled and the motions on XYZ axis and the yaw motion are controlled using PD (Proportional- Derivative) controllers. In [11], a control design, using the root locus analysis, was developed to solve the tracking prob- lem on the XY frame and the pole placement technique based on assigned dominant closed loop pole and a velocity feed- back have been used. In [12] an attitude control is applied to a quadrotor heli- copter taking into account artificial visual and IMU (Inertial Measurement Unit) sensor. The translational and rotational movements on the Z axis were modeled as integrators and were controlled by proportional control laws. The transla- tional motions on the X and Y axes are controlled by PD con- trollers, giving the system a better phase margin. In [13], ar- tificial visual and IMU sensor are also used to develop a filter to estimate the attitude of a four rotor helicopter. Firstly, at each vision step the coordinates of four LEDs (Light-Emitting Diode) attached to the helicopter body are measured. Then, these measurements are converted into the Euler angles esti- mator by solving a nonlinear least squares problem. Secondly, at each gyro step, the measurements of body axis angular ve- locity are transformed into Euler rates, using standard kine- matic equations. Finally, a Kalman filter is used to merge the two different sources and compute an optimal estimation of Euler angles. This paper is derived from a benchmark organized by CEA (Comité Español de Automática), whose aim is to design a control law to perform tracking on the XY plane for au- 1