Extraction and Implementation of Cooperative Behavioral Decision Rules of Human into Robot by Using Tele-operation System with Immersive Vision

为了建立自主式移动的机器人的行为决策程序，人们采用了基于动力学、几何学、统计学等的各种算法。然而，有几个困难，开发这样的算法。本项目作为自主机器人程序的另一种方法，研究了人类合作决策规则的提取及其在机器人中的实现，并为此开发了沉浸式远程控制系统。该系统可以给操作者虚拟的机器人的物理特征。将机器人上的全方位摄像机拍摄的图像转换为透视图像，并投影到操作者周围的三个屏幕上。这使得操作员能够获得机器人周围的视觉信息，就像他/她骑在机器人上一样。操作者可以通过操纵杆向机器人发出指令，利用该系统进行了多项实验，提取了人类运动决策规则。从人的操作分析，我们把重点放在地板面积投影到每个屏幕上的环境信息。这些外部信息与人的行为（操纵杆的输入）之间的对应关系被定义为“人的决策规则”。由于人的操作具有非连续性和非线性，很难用单一的函数来表示。因此，将人的行为划分为若干操作模式，并将对应的关系存储为数据库。在实现了该对应关系数据库后，机器人的运动表现出了一些类似于人操作机器人的运动模式。

In order to build programs of behavioral decision for autonomous mobile robots, various type of algorithms based on dynamics, geometry, statistics and so on have been adopted. However, there are several difficulties to develop such algorithms. In this project, as another approach to building programs for autonomous robots, the extraction of cooperative human decision rules and the implementation of them into a robot are studied.As preparation for that, the immersive remote control system has developed. This system can give an operator the physical feature of the robot virtually. Image taken with omni-directional camera on the robot is transformed into perspective image and is projected onto the three screens around the operator. This enables the operator to get visual information around the robot as if he/she rides on the robot. The operator can send commands to the robot by using a joystick.By using this system, several experiments are carried out to extract human decision rules moving action. From the analysis of human operation, we focus on the floor area projected onto the each screens as environmental information. The corresponding relation between these external information and human behaviors (input for the joystick) is defined as "human decision rules". Because of discontinuity and nonlinearity of human operation, it is so difficult to represent human operation by a single function. Therefore, human behavior is divided into some operation patterns and the corresponding relations are stored as database. After the implementing of this database of corresponding relations, the motions of the robot show some patterns which are similar to those of the robot operated by human.