A study of learning model of motor command and symbol for communication

关于人类手臂在平面上的点对点运动的特征，已经很好地演示了以下两个特征。已经提出了几个模型来解释这些特征。提出了基于最优原理的四个准则。在该项目中，我们测量了三维空间中点对点的运动，并将最优准则规划的手臂姿势与测量的手臂姿势进行了定量的对比检验。接下来，基于最小角加速度准则、最小扭矩变化准则和最小指令扭矩变化准则，对测量轨迹和最优轨迹进行了对比检验。最后，我们报告了测量的手部轨迹和手臂姿势与最小指令扭矩变化准则预测的轨迹和姿势最接近，人类可以使用复杂的动作进行交流。然而，所提出的模型的重点一直是从运动轨迹到表征的转换以及从表征再生运动。换言之，这些模型不能模拟包含大量试错的人类表征获取过程。我们研究了复杂运动模式和符号的需求过程。我们提出了两种模型，一种是估计运动的随机表示(如通孔)的模型，另一种是估计的表示与符号相关的模型。第一种模型利用迭代过程获得一组通孔，从而平滑地重新生成原始图案。在学习实验中，我们试图获得通道点表征中的运动模式。实验结果表明，我们的模型能够获得使其能够重新生成示例模式的表示。我们确认我们的第二个模型可以与符号和图案连接。这些模型可以为沟通提供一个基本的模型。

The following two characteristics have been well demonstrated about the feature of a point-to-point human arm movement on a plane. Several models have been proposed to explain the features. Four criteria based on optimal principles have been proposed. In the project, we measured point-to-point movements in three-dimensional space and performed comparative examinations quantitatively between arm posture planned by the optimal criterion and measuerd arm posture. Next, we performed comparative examinations between the measured trajectories and the optimal trajectories based on minimum angle jerk criterion, minimum torque change criterion and minimum commanded torque change criterion. Finally, we report that both of the measured hand trajectory and arm posture were closest to the trajectory and posture predicted by the minimum commanded torque change criterion.Humans can communicate using complex movements. However, the focus of the proposed models has been on transformation from movement trajectories to representations and on the regeneration of movements from representations. In other words, the models can not emulate the human representation acquirement process which includes lot of trial and error. We study a requirement process for complex movement patterns and symbols. We propose two models, one is a model that estimates the stochastic representation of a movement such as via-points, and the other is a model in which an estimated representation is related to a symbol. The first proposed model can acquire a set of via-points to regenerate an original pattern smoothly using on iterative process. In learning experiments, we tried to acquire movement patterns in via-point representation. As the result of experiments, we confirmed that our model can acquire representation that enables it to regenerate the example patterns. We confirmed that our second model can connect with symbols and patterns. The models can provide a fundamental model for communication.