Input and plant parameter optimization via learning optimal control of Hamiltonian systems

The proposed method solves a class of optimal control problems based on a symmetric property of Hamiltonian systems. It allows one to simultaneously obtain an optimal feedforward input and optimal plant parameters, which (at least locally) minimize a given cost function. The big advantage of this method is that the precise model of the dynamics of the plant system is not required by using the symmetric property of Hamiltonian systems, called variational symmetry. We also show some applications of this method to optimal gait generation for a one-legged hopping robot.