公开(公告)号：KR1020070072314A

公开(公告)日：2007-07-04

申请号：KR1020060040506

申请日：2006-05-04

Applicant: 고려대학교 산학협력단

Inventor： 박귀태 ,  김동원

IPC: B25J13/08 ,  B25J5/00

Abstract: A robot posture control method using a neural network, a recording medium thereof, a robot posture control apparatus using a neural network, and a robot having the apparatus are provided to allow the robot to stably walk even under a new environment or when a mechanical error exists. A robot posture control method using a neural network includes a step(1610) of measuring a balancing point by using the pressure applied to a pressure sensor during operation of a robot performed in accordance with a predetermined walking pattern value; a step(1620) of creating a balancing point error from the difference between the measured balance point and a pre-stored balancing point, and outputting the balancing point error to a predetermined neural network; a step(1630) of creating a joint angle adjustment value by assigning weight values of the neurons contained in the neural network to the balancing point error; a step(1640) of operating an actuator in accordance with the control signal obtained by summing the joint value adjustment value and the walking pattern value; and a step(1650) of varying the weight values by performing a back propagation algorithm in accordance with the balancing point error created during operation of the actuator.

Abstract translation: 使用神经网络的机器人姿态控制方法，其记录介质，使用神经网络的机器人姿势控制装置和具有该装置的机器人，以使机器人即使在新的环境下也能稳定地行走，或者机械误差 存在。 使用神经网络的机器人姿势控制方法包括：步骤（1610），其通过使用根据预定行走模式值进行的机器人的操作期间施加到压力传感器的压力来测量平衡点; 从所述测量平衡点和预先存储的平衡点之间的差产生平衡点误差的步骤（1620），并将平衡点误差输出到预定的神经网络; 通过将神经网络中包含的神经元的权重值分配给平衡点误差来创建关节角度调整值的步骤（1630） 根据通过对联合值调整值和行走模式值进行求和而获得的控制信号来操作致动器的步骤（1640） 以及通过根据在致动器的操作期间产生的平衡点误差执行反向传播算法来改变重量值的步骤（1650）。