Tele-operation of service robots by non-Japanese interactive teaching

本研究试图为远程机器人建立一个人性化的界面。这里考虑的情况如下。来自海外的非日本人控制机器人，用于照顾日本老人。在这种情况下，操作人员与日本人有着完全不同的文化背景，因此要实现这种远程护理，非常人性化的界面是绝对必要的。本研究采用两种方法：1)口头交流：简单的口头句子在翻译上可能没有问题。因此，我们尝试通过简单的日常语言交流来控制远程机器人。2)物理运动：物理运动本身具有一定的意义，因此我们试图从操作者的运动中找回一些意义。在第一种方法中，我们把句子中的名词作为最重要的项目。操作员不用说日常会话，而是用鼠标在屏幕上指示操作员计划在会话中说的对象（名词）。开发的接口重建了几个合适的候选指令，这些指令可能是操作员意图的机器人操作命令。通过选择其中一个，所需的任务将被执行。本研究还开发了一种实用的方法来操纵远程操作员所指示的对象。关键是要知道物体的真实位置和处理方法。机器视觉技术和数据库技术解决了这些问题。在第二种方法中，我们成功地识别了操作员移动手臂时的基本意图。也就是说，我们可以区分手臂运动中包含的三种不同的意图：触摸，撞击或推动。这种方法对于通信通道中的延迟非常健壮。

This study tried to build a human friendly interface for tele-robots. The situation considered here is as follows. Non-Japanese control robots located in Japan from overseas for giving cares to Japanese elderly. In such case, the operators have complete different cultural background from Japanese, thus quite user-friendly interfaces are absolutely necessary to realize such tele-caring. This study took two approaches :1) Verbal Communication : Simple verbal sentences may have no problems to interpret. Thus, we tried to control the remote robot by simple daily verbal communication.2) Physical Motion : Physical motion itself has some meanings, thus we tried to retrieve some meaning from the operators' motion.In the first approach, we treat nouns in the sentence as the most important items. In place of saying daily conversation, operators indicate the objects(nouns) in the screen by computer mouse, which the operator planned to say in his/her conversation. The developed interface rebuilds several appropriate candidates that might be robot-operation commands of operators' intention. By choosing one of them, the desired task will be executed. This study also developed a practical way to manipulate the object indicated by the remote operators. The crucial points are to know the real location of the object and the way to handle it. A machine vision technique and data-base technique resolved these problems. In the second approach, we succeeded to identify the basic intention of operators when they move their arms. Namely we could distinguish three different intention included in the arm motion : touch, hit or push. This approach is quite robust against delays in communication channels.