Development of a sequential learning and control scheme for a quadrotor altitude model

The contribution of this thesis lies in the field of system identification techniques, which require estimating the parameters of a dynamical system from data collected during experiments. This is essential in many fields of engineering since, in real-life experiments, the perfect knowledge of the coefficients related to the equations describing the dynamics of the system is not always available. Specifically, this thesis investigates a recent system identification technique in the presence of non-stochastic energy-bounded disturbances, to obtain a desired error bound on the estimated parameters. The first objective of this thesis lies in developing a Software In the Loop (SIL) simulator to test this technique based on optimization problems, on a quadrotor model. This SIL simulator is based on an existing simulator (RotorS developed by ETH Zurich) running on ROS infrastructure, adapted to collaborate with MATLAB where the optimization techniques are implemented. Moreover, the quadrotor model is derived, and a pre-stabilizing controller is implemented to conduct the desired experiments without accounting for the unstable nature of the quadrotor. Finally, the techniques to generate the targeted exploration signal for the system identification are implemented and tested only on the altitude model of the quadrotor. The results of the simulation show that the exploration signal generated by the optimization techniques is able to reduce the uncertainty of the model and achieve the desired accuracy. This thesis is a first step toward implementing a robust control strategy based on the desired uncertainty obtained from the system identification techniques known in the literature as dual control.