Abstract
This paper concerns an optimal gait generation with respect to energy consumption by learning trajectory and adjusting robot parameters based on learning optimal control. In this method, learning optimal control of Hamiltonian systems, which unifies learning control and parameter tuning, plays a key role. It allows one to simultaneously obtain an optimal trajectory and tuning parameters for a plant system, which (at least locally) minimize a given cost function. The proposed method is applied to the compass gait biped on a shallow slope and the one with a torso on the level ground, respectively. Consequently, a passive dynamic walking is generated for the first case, and an energy-efficient walking trajectory is generated for the latter case.
