基于神经网络的空间7R冗余机器人的运动模型辨识

Kinematic Model Identification of Spatial 7R Redundant Robot Based on Neural Network

针对空间冗余机器人运动学控制中正、逆运动学求解的复杂性,采用神经网络从两方面解决这一问题.一是从神经网络出发,提出了一种新的动态神经网络结构——状态延迟输入动态递归神经网络(SDIDRNN),提高了网络的学习速度;二是从辨识方案出发,以SDIDRNN为基础,在空间7R冗余机器人正、逆运动学模型辨识的问题上,设计了一种新颖的解耦辨识方案.将其与另外两种具有普通网络结构的辨识方案相比较,说明了该新方案具有更高的学习能力,辨识误差可降低到对比方案的40％-6％.由于学习速度的提高,达到设定误差时的训练次数大大减少,使该方案在机器人运动控制系统中的实时计算能力大大增强,为神经网络在机器人运动学控制中的应用提供了一条崭新的思路,具有重要的应用意义.

Neural network is used in this paper to solve the problem of kinematics and inverse kinematics in kinematical control of spatial redundant robots. Two methods are explored to increase computational efficiency of neural network. First, a new neural network model named State Delay Input Dynamical Recurrent Neural Network ( SDIDRNN) is proposed on the basis of Elman network to improve learning rate and static accuracy. Second, a novel identification scheme is designed. Based on SDIDRNN, the new decoupling identification scheme is used for kinematic identification of a spatial 7R redundant robot. It is compared with another two identification schemes following normal identification idea. Simulation results show the considerable improvement of learning ability of the new scheme. After being trained some times (10 times) , the root-mean-square errors of the new scheme decrease to 40% -6% of those of the other two schemes. This superiority makes the possibility of online identification or computation increase greatly in kinematical control of spatial redundant robot systems.