Neural Network Model for Voluntary Movement and Application to Robotics

人类的运动技能不是天生的，而是从出生起就必须经过训练才能获得的。一开始，运动是通过视觉通路的反馈信号来控制的。随着技能的提高，反馈控制系统将被前馈控制系统所取代，这意味着可以无意识地控制运动。提出了一种可以解释这一过程的神经网络模型。该模型基于一种被称为反馈-错误-学习的规则。根据反向传播学习规则将运动系统的逆动力学组织在一个三层神经网络中。讨论了人体手臂运动的最优控制，提出了基于最小转矩变化准则实现最优路径的神经网络模型。基本思想是(1)时间的空间表示；(2)正动力学和运动学模型的学习；(3)基于获得的模型进行松弛计算。该网络的操作分为学习阶段和模式生成阶段。在学习阶段，该网络获取多自由度被控对象的正演模型，同时监测实际轨迹作为教学信号。在模式生成阶段，激活代表相邻时间运动指令的神经元之间的电耦合，以保证最小转矩变化准则。计算机仿真结果表明，该模型能够在避障和通过过点时产生多关节臂轨迹。

Human motor skills are not innate, but have to be acquired by training from birth. In the beginning, movements are controlled by using feedback signals through visual pathways. With increased skill, the feedback control system is replaced as the main controller by a feedforward control system which means movements can be control led unconsciously. A neural network model which can explain this process is proposed. The model is based on a rule called the feedback-error-learning. The inverse dynamics of the motor system is organized in a three layer neural network according to the back-propagation learning rule.Optimal control of human arm movement is also discussed , and a neural network model which realizes the optimal pathways based on the minimum torque change criterion is proposed. Basic ideas are (1) spatial representation of time, (2) learning of forward dynamics and kinematics model and (3) relaxation computation based on the acquired model. Operations of this network are divided into the learning phase and the pattern-generating phase. In the learning phase, this network acquires a forward model of the multi-degree-of-freedom controlled object while monitoring the actual trajectory as a teaching signal. In the pattern-generating phase, electrical coupling between neurons representing motor commands at neighboring times is activated to guarantee the minimum torque-change criterion. By computer simulation, we show that the model can produce a multi-joint arm trajectory while avoiding obstacles or passing through viapoints.