Adaptive Control in the Presence of Sensor Measure Outliers

Adaptive control of plants whose sensor measurements are corrupted by outliers is considered. Outliers are large deviations of the signal being measured, only occurring in a few percent of the observations. For adaptive controllers relying on an implicit Gaussian assumption, both the identification and the underlying control law are yielded by the minimization of quadratic losses. Therefore, although rare, outliers heavily degrade performance due to their large amplitude. This problem is tackled in this paper. Methods for outlier removal which are suitable for adaptive control applications are reviewed and illustrated through an application to position control in the ball and beam plant. 1