Robot Arm Touch Detection by Signal Separation with Higher Order Statistics and Compliance Control

本研究的目的有两个方面：1)利用高阶统计量的信号分离来完成控制系统的分析方法。统计学习中的一种新方法，称为独立分量分析，使我们能够分析基于高阶统计量（如峰度）的信号。从而可以从混合信号中分离出独立分量，恢复源信号。在本研究中，我们改进了这种方法，并开发了一种分析和综合控制系统的方法。2)推导出机器人手臂对障碍物的触摸检测机构。通过实时检测，我们可以停止或放松手臂运动。此外，我们的目标是开发一种合规控制技术。关于1)，我们已经完成了该方法，并将其应用于控制中的一些问题。然后，通过反转分离与观测信号逆操作的思想，我们开发了所谓的反馈误差学习（FEL）的多变量版本，其中命令输入的前馈控制器通过一些学习规律进行调整，以获得更高的运动精度。前一个结果，我们获得了SICE（我们的主要学会）的最佳论文奖，后一个结果，我们获得了SICE科学技术部的最佳演讲奖。另一方面，我们还没有在这个项目上取得成功，部分原因是我们把注意力集中在上述主题上。但是我们已经成功地用机械臂写了一笔字，这是第2点的副作用。我们正在继续努力解决这个问题。

The objectives of this research are two-folds:1) To complete an analysis method for control systems by means of signal separation with higher order statistics. A novel method in statistical learning, named independent component analysis, enables us to analyze signals based on higher order statistics such as Kurtosis. Thereby we can separate independent components from the mixture of signals and recover the source signal. In this research, we modify this method and develop a method for analysis and synthesis of control systems.2) To derive a mechanism for touch detection of a robot arm to some obstacle. By real-time detection we can stop or relax the arm motion. Furthermore, we aim at developing a compliance control technique.Concerning 1),we have completed the method and applied it to some problems in control. Then, by reversing the idea of separation with inverse operation of observed signals, we have developed multivariable version of what is called Feedback Error Learning (FEL), where a feedforward controller for command inputs is adjusted by some learning law to obtain higher accuracy of the motion. For the former result, we have obtained a best paper award from SICE (our main academic society), and for the latter result, we have obtained a best presentation award from SI division of SICE.On the other hand, we have not yet succeeded in the item2), partly because we have concentrated on the above topics. But we have succeeded in writing one-stroke characters by a robot arm, which is a side-effect from 2). We are continuing efforts on this issue.