CMAC神经网络已被广泛用于机械手的控制 ,但这种网络的连接权矩阵通常由学习得到 ,有时因学习时间长而影响控制的适时性。本文提出一种连接权矩阵的计算方法 ,不用学习可直接计算求得连接权矩阵 ,其均方根误差近似为零 ,从而实现了用于机械手控制的机械手 CMAC神经网络的记忆。保证了机械手控制的适时性。 CMAC neural network has been widely used in manipulator control, but the connection matrix of this kind of network is usually obtained by learning, and sometimes due to long learning time affect the control of timeliness. In this paper, we propose a method to calculate the connection matrix, which can be directly calculated without learning to obtain the matrix of connection weights. The root mean square error of the matrix is ​​approximately zero, so that the CMAC neural network memory for manipulator control is realized. Ensure the robot control of the timeliness.