Incremental Nonlinear Dynamic Inversion with Sparse Online Gaussian Processes Adaptation for Partially Unknown Systems

Abstract

Sensor-based Incremental control is a recently developed family of techniques with a reduced dependency on a plant model. This approach uses measurements or estimates of current state derivatives and actuator states to linearize the dynamics with respect to the previous time moment. However, in such a formulation, the control system is sensitive to the quality of measurements or estimations. The presence of uncertainties caused by unforeseen malfunctions in measurement and/or actuation systems could provoke drastic performance degradation. The paper proposes a sensor-based Incremental Nonlinear Dynamic Inversion (INDI) control algorithm augmented with Budgeted Sparse Online Gaussian Processes Adaptation for the compensation of unknown system behaviour. INDI performs quite efficiently under design conditions. Meanwhile, GP-based direct adaptation provides not only long-term dependency learning but also noise signal filtering. The efficiency of the proposed approach is demonstrated with a longitudinal motion of a missile.