Eva Ahbe PhD Researcher
ETH Zurich Automatic Control Laboratory
Physikstrasse3 8092 Zurich Switzerland
Experimental Validation of Path-Tracking Model Predictive
Control for Fixed-Wing Power Kites
Eva Ahbe1, Thomas Stastny2, Manuel Dangel2, Roland Siegwart2, Roy S. Smith1 1Automatic Control Laboratory, ETH Zurich
2Autonomous Systems Laboratory, ETH Zurich
Model Predictive Control (MPC) has previously been em-ployed in simulations for the control of flexible as well as rigid wing AWE systems . Successful experimental vali-dation of MPC based controllers have been reported from tow-test experiments with flexible wing kites . Though MPC offers a promising approach to the control of AWE systems through the advantage of incorporating explicit models of the system and maximize for performance, the downsides of MPC such as model mismatch, real-time operational requirements, and solver feasibility depen-dence are still posing major challenges to the real-world application of MPC in the field of airborne wind energy. In this talk we present experimental validation of the autonomous flight of a tethered Easy Glider system us-ing a path-trackus-ing nonlinear MPC algorithm. The proof of concept is hereby provided for the system flying in crosswind-like conditions, while a propulsion on the ve-hicle is used to guarantee a minimum flight velocity. The MPC controller is designed to follow a circle-shaped ref-erence path. This refref-erence path is obtained as a func-tion of the mass, lift coefficient and tether length with the aim of minimizing the aerodynamic losses due to steer-ing. The MPC algorithm takes constraints into account which consist in state (i.e. height and velocity) and input constraints. The model of the system used by the MPC consists of a nonlinear unicycle-like model with an addi-tional state in form of the kite velocity.
Flight experiments of the proposed control scheme were
performed with a 1.8 m wingspan, 1.7 kg, foam Easyglider test platform. The glider was installed with a Pixhawk Flight Controller  168 MHz Cortex-M4F microcontroller with 192 kB RAM. Additionally, an on-board companion computer, Intel Up Board (Quad Core, 1.92 GHz CPU, 4 GB RAM), was installed running Robotic Operating System (ROS) for generating MPC solutions in real time and trans-mitting attitude references to the Pixhawk. The experi-mental flight test shows the autonomous flight of circu-lar trajectories close to the reference trajectory. The re-maining discrepancy between reference and actual path is mainly due to the low tether force due to low wind speed conditions. Additionally, there was a significant model mismatch introduced by the usage of an unmod-eled tether of relatively large mass. Future research will focus on improved system identification, including a tether model and further tests under higher wind condi-tions.
 S. Gros, M. Zanon, and M. Diehl, łControl of Airborne Wind Energy systems based on Nonlinear Model Predictive Control and Moving Horizon Estimation,ž in 2013 Eur. Control Conf., pp. 1017ś1022, IEEE (2013)
 T. A. Wood, H. Hesse, and R. S. Smith, łPredictive Control of Au-tonomous Kites in Tow Test Experiments,ž IEEE Control Syst. Lett., vol. 1, no. 1, pp. 110ś115 (2017)