103 ニューラルネットワークを用いたエネルギ最少歩行のためのロボット脚の軌道生成

抄録

In this paper, we study an inverse kinematic problem in which trajectories of walking robot toes are generated under the condition of minimum walking energy. Under spatial constraints such as obstacle avoidance, mathematical optimization method is used for the optimization of several parameters which describe the trajectory function of toe. The calculation of optimum trajectory takes much time, therefore it is difficult to control for walking robot in real time. In this study we use a neural network controller. Data of optimum trajectory obtained by using the mathematical optimization is used for learning of the neural network. Under spatial constraints such as an obstacle, the optimum trajectories of toes are generated by the neural network controller and the controlling values of joint angles are successfully obtained.