Model Predictive Control techniques for fixed-wing UAV maneuvers

The most common control algorithm used in commercial Unmanned Aerial Vehicles’ (UAVs) autopilots is the Proportional Integrative Derivative controller. The PID-based controller cannot handle changes in UAV dynamics and the presence of wind disturbances; in this latter case a PID parameters re-tuning is necessary. UAVs’ flight performance is affected by exogenous disturbances and additive noise, existing in a real operative environment. Dealing with always more demanding requirements of flight maneuvers, a robust Model Predictive Control (MPC) approach is proposed, which is able to handle external disturbances (as gusts or wind distubances) and parametric uncertainties (as variations in mass, flight conditions or payload). In this work, it is first considered a classical MPC design. A cost function and state and control constraints are built for the inner loop dynamics (pitch, roll and airspeed) and for the altitude outer loop. A PID control regulates the heading variation (navigation outer loop). MPC control law is based on an optimization problem, which at each sampling time aims to find the optimal control sequence that minimizes some function and satisfies some constraints; only the first element of the predicted optimal control sequence is applied to the plant. To obtain a feasible problem it is necessary to choose accurately tuning parameters. An interesting variation of classical MPC is the Tube-based Robust MPC (TBRMPC), which lets to deal with external bounded disturbances and parametric uncertainties with the same computational efficiency of a classical MPC and guarantees to respect hard constraints. In the TBRMPC, a linear nominal system is taken into account and it represents a nominal dynamics of the system with no disturbances and uncertainties. The discrepancy between nominal and actual system lets to define the error dynamics. The TBRMPC algorithm consists of an offline part and an online part. In the offline part a feedback gain matrix is evaluated in order to stabilize error dynamics. The advantages of using this kind of controller are twofold: (i) low computational effort, so this controller can be implemented on a on-board controller, (ii) guarantee of robustness of the control system, able to handle variations of the system and to represent a realistic environment. The key feature of this proposed approach is the real-time implementability, with a time-varying control law ans, as said before, a feedback gain evaluated offline. Moreover, tighted state and control constraints are computed. In the online part a classical MPC optimization problem is solved at each sampling time, a nominal input is derived and then corrected according to the gain feedback matrix and the actual error. A Linear Matrix Inequality (LMI) approach is applied to the state feedback stabilization, to reduce the computational effort, guaranteeing the stability and improving real-time implementability.