Enhanced LQR Control for Unmanned Helicopter in Hover Zhe Jiang, Jianda Han ,Yuechao Wang and Qi Song Abstract-Real time adaptability is of central importance for linearization and optimization", but the implementation on the control of Unmanned Helicopter flying under different real system was constrained by the non-minimum phase circumstances. In this paper, an active model is employed to characteristics, the computational complexity as well as the handle the time varying uncertainties involved in the helicopter lack of robustness [3]. On the other hand, robust control, such dynamics during flight. In the scheme, a normal LQR control designed from a simplified model at hovering is enhanced by as H-nfinity control [5], is another algorthm proposed for means of Unscented-Kalman-Filter (UKF) based estimation, helicopter, but it Is difficult to make an optimal balance which tries to online capture the error between the simplified between robustness and conservation, which usually leads to model and the full dynamics. This is intended to achieve an over- conservative controller to guarantee robustness. adaptive performance without the need of adjusting the Besides the control algorithms explicitly based on off-line controller modes or parameters along with the changing model, online modeling, which tries to build a high fidelity dynamics of helicopter. Simulations with respect to a model model on board and in real time, becomes an important helicopter are conducted to verify both the UKF-based ' estimation and the enhanced LQR control. Results are also direction for the high precision control of vehicles with demonstrated with the normal LQR control with the active nonlinear and time-varying dynamics. In [3], the algorithm model enhancement. known as active model estimation plays an important role in autonomous control architecture because it provides I. INTRODUCTION necessary information required by robust control, trajectory T he Dynamics of an unmanned helicopter is strongly generation, mission planning, and control reconfiguration for nonlinear, inherently unstable, highly coupled and forms fault accommodation, to the autonomous system that adapts a multiple input multiple output (MIMO) non-minimum over time. The information which an active model can phase system with time varying parameters. At the same time, provide includes [6]: a) state estimation, which provides the dynamics is also influenced by the turbulence from tail necessary information for full state feedback control; b) rotor and lateral wind. A helicopter also has multiple flight predictive model, which builds online model for adaptive modes, such as hovering, forward, backward, sideslip, up and control and fault detection and identification; and c) downward flights as well as other aggressive maneuvers uncertainty bound, which forms the basis of robust control. combining the basic patterns. A helicopter has 6 degree-of- Neural Networks (NN) and NN-based self learning have freedom (DOF), but it is usually controlled by 4 actuator been proposed as one of the most important approaches for inputs, which means that the control of each DOF is coupled active modeling of unmanned vehicle [7-9]. However, the but independent. All of the above mentioned issues problems involved in NN, such as training data selection, complicate the helicopter dynamics and make it impossible to online convergence, robustness, reliability and real-time build high fidelity model for a helicopter [1]. implementation, limit its extensive application in real Helicopter flight control system design has been systems. dominated by classical control techniques. But, recent years Most recently, researches are focusing on the sequential have seen a growing interest in applications of nonlinear estimation and its applications on active modeling and control theory when a nonlinear model is deployed for the model-reference control [10]. The classical state estimator for controller design. This is mainly due to the increasing nonlinear system is the extended Kalman Filter (EKF). attempts on making a helicopter to be unmanned and Although widely used, EKF's have some deficiencies, mainly autonomous. Feedback linearization [2] and State Dependent due to its linearizing the nonlinear dynamics. The UKF, on Riccati Equation (SDRE) method [4] tried to handle the the other hand, has the same computational complexity with nonlinear dynamics of a helicopter by the way of "on-line the EKF, but directly use the nonlinear models instead of linearizing it. UKF does not need to calculate Jacobians or Manuscript received December 9, 2005. This work was partially Hessians and can achieve the second-order accuracy (the supported by the National High Technology Research and Development accuracy of EKF is first-order). Therefore, UKF is well suited Program under the grant 2003AA421020. 7 T;,,---- -41, 1,- 'D-1-+;-, Tfor online application in nonlinear systems with fast Z. Jiang was with the Robotics Laboratory, Shenyang Institute of fo onin aplcto. nnniea ytm ihfs Automation, Chinese Academy of Sciences, Nanta street 1 14#, Shenyang, time-varying dynamics. 110016, China. He is now with the Graduate School, Chinese Academy of In this paper, a normal LQR control designed from a Sciences, Beijing 100864, China (phone: ±86-024-23970721; fax: smlfe molathvrn isnacdby ens f ±86-024-2397002 1; e-mail: zha (~ic). J.D. Han, Y.C. Wang and Q. Song are with the Robotics Laboratory, Unscented-Kalman-Filter (UKF) based estimation, which Shenyang Institute of Automation, Chinese Academy of Sciences, tries to online capture the error between the simplified model China.(e-mail: idhanKX7sin, Xcwa ~siac, son im~sacn) 1438