Research on a Robot that Operates Based on a Dialog with a User.

随着我们面临老龄化社会，对这种能够在各种日常情况下帮助人类的个人服务机器人的需求越来越大。我们正在开发一种个人服务机器人的原型，它能够从远处的冰箱里取罐头。我们对这类服务机器人所需的以下功能进行了研究：移动机械手的易示教：通常服务机器人要处理比工业机器人更广泛的任务。因此，对于这样的服务机器人来说，一种简单、用户友好的教学方法是可取的。我们开发了以下教学方法。用户向机器人传授手的标称轨迹及其容差以完成任务。机器人在给定的公差内搜索可行的轨迹；如果需要，它还会产生移动基座的运动。由于用户只需要考虑手的运动，该方法很直观，不需要用户太多的努力；另外，由于给出了名义轨迹，机器人不需要太多的计算来生成轨迹。交互式对象识别：机器人基本上利用颜色信息识别罐头和瓶子。如果机器人未能找到目标对象，它会尝试通过与用户对话来获取其他信息。机器人生成一个很好的问题，可以从用户那里检索到信息丰富的答案；这是机器人和用户之间顺利交互的关键。Robust语音识别：许多现有的基于对话的界面系统假设语音识别(SUB)系统总是工作良好。然而，由于与机器人的对话通常是在存在各种噪声的环境中进行的，因此不能做出这样的假设。我们已经开发了一种从(错误的)识别的单词、识别的单词之前和之后的识别的单词以及对话的状态来估计未识别的单词的含义的方法。

As we are facing the "aging society", the need for such personal service robots which can help human in various everyday situations is increasing. We are developing a prototype of personal service robot that is able to fetch a can from a distant refrigerator. We have performed the research on the following functions required for such service robots:Easy teaching of mobile manipulator: Usually service robots have to deal with much wider range of tasks than industrial ones. An easy, user-friendly teaching method is, therefore, desirable for such service robots. We developed the following teaching method. A user teaches the robot a nominal trajectory of the hand and its tolerance to achieve a task. The robot searches for a feasible trajectory within the given tolerance; it also generates the movement of the mobile base if necessary. Since the user only needs to consider the movement of the hand, the method is well intuitive and does not require much effort from a user; in addition, the robot does not need much computation for trajectory generation because a nominal trajectory is given.Interactive object recognition: The robot basically recognizes cans and bottles using color information. If the robot failed to find a target object, it tries to obtain additional information by a dialog with a user. The robot generates a good question which can retrive informative answers from the user; this is a key to a smooth interaction between the robot aud the user.Robust speech recognition: Many existing dialog-based interface systems assume that a speech recognition (sub)system always works well. However, since the dialog with a robot is usually held in environments where various noises exist, such an assumption cannot be made. We have developed a method for estimating the meaning of unidentified words from the (mistakenly) recognized words, the identified words before and after the unidentified one, and the state of the dialog.