Optimal control of Hamiltonian systems via iterative learning

This paper is concerned with optimal control of Hamiltonian systems with input constraints via iterative learning algorithm. The proposed method is based on the self-adjoint property of the variational systems of Hamiltonian systems. This fact allows one to execute the numerical iterative algorithm to solve optimal control without using the precise model of the plant system to be controlled. A learning framework for an optimal control problem to achieve a prescribed desired terminal state under input saturations is proposed. A concrete learning algorithm for mechanical systems is also derived. Furthermore, numerical simulations of a 2-link robot manipulator demonstrate the effectiveness of the proposed method.