Iterative Learning Optimal Control of Hamiltonian Systems Based on Variational Symmetry

Abstract

This paper proposes a novel iterative learning control method for Hamiltonian control systems which can solve a class of optimal control problems. First of all, a symmetric property of the input-output mappings of Hamiltonian systems is clarified which plays an important role in solving, optimal control problems by gradient method. A concrete learning algorithm is derived for mechanical systems possibly with input saturation. Furthermore, numerical simulations of a 2 link robot manipulator demonstrate the the effectiveness of the proposed method.