Q-learning-based feedback linearization method for unknown dynamics

Abstract

This paper present a feedback linearization technique for affine nonlinear systems that is independent of system dynamics. First, a input-output feedback linearization correction framework is described, and a interference estimator is employed to guarantee the stability of plant during the learning process. Then, a model-free Q-learning algorithm is presented to solve the feedback linearized controller. Finally, the position control of a single-link flexible joint manipulator system is used as an example to demonstrate the effectiveness of the method.