Quadcopter flight test results of a consistency monitoring algorithm for adaptive controllers

Model Reference Adaptive Control facilitates nonlinear systems to adapt to modeling errors, environmental changes or structural damage. Most adaptive control frameworks employ Lyapunov analysis in order to establish stability of the closed loop system. However, the derived signal bounds are inherently conservative. Furthermore, these bounds are seldom known; the plant states might still violate structural requirements, thus causing the system to fail. In order to counter this disadvantage, we employ an adaptive output consistency monitor, which is based on Bayesian linear regression. It learns a model of the adaptive controller from online gathered data and thus allows for the assertion whether the adaptive output is within an interval of the expected output. The method is demonstrated in flight tests of a quadcopter with and without robustness modification.