A Study on Human-Robot Interface based on Uncertainty Modeling and Inference

研究的目的是研究人机接口，它考虑到各种不确定性的任务环境和人为因素。我们开发了两个系统：（1）人工任务。引导系统：为了引导人类任务，移动的机器人必须科普环境和机器人传感器中的各种不确定性。针对移动的机器人定位问题，提出了一种传感器规划方法。我们表示的因果关系，局部感知结果，行动，和信念的全球定位使用贝叶斯网络。首先，贝叶斯网络的结构学习从环境的完整数据使用K2算法结合遗传算法（GA）。在执行阶段，当机器人被绑架到某个地方，它计划一个最佳的传感行动，考虑到传感成本和全局定位信念之间的权衡，这是通过推理得到的贝叶斯网络。我们已经验证了学习和规划算法的仿真和真实的机器人实验在办公室环境中。(2)导航辅助系统：我们开发了一种智能轮椅，它允许我们使用两种方法来指导它的运动，一种是使用手（和手指）手势，另一种是使用语音控制。第一种方法使用红外摄像机来可靠地提取图像中的手区域。图像处理器提取该区域的主轴作为手的方向。第二种方法是使用名为Julius的语音识别子系统开发的，该子系统使用隐马尔可夫模型来识别引导运动的单词，例如日语中发音的“Mae”，“Migi”，“Hidari”，“Tomare”等。通过使用一些标准的识别结果，我们已经进行了实验，以验证使用语音命令以及手势命令来正确控制轮椅的人机界面。

Purpose of the research is to investigate human-robot interface, which takes into account various uncertainties contained in the task environment and human factors. We have developed two systems:(1) A human task. guiding system:To guide human-tasks, a mobile robot has to cope with various uncertainties in the environment and the robot sensors. We developed a method of sensor planning for a mobile robot localization problem. We represent causal relation between local sensing results, actions, and belief of the global localization using a Bayesian network. Initially, the structure of the Bayesian network is learned from the complete data of the environment using K2 algorithm combined with GA (genetic algorithm). In the execution phase, when the robot is kidnapped to some place, it plans an optimal sensing action by taking into account the trade-off between the sensing cost and the global localization belief, which is obtained by inference in the Bayesian network. We have validated the learning and planning algorithm by simulation and real robot experiments in an office environment.(2) Navigation assisting system:We have developed an intelligent wheel chair, which allows us to direct its movement using two methods, one to use hand (and finger) gesture and the other to use voice control. The first method uses an infrared camera to reliably extract hand region in the image. An image processor extracts principal axis of the region, as the hand direction. The second method is developed using a voce recognition subsystem named Julius which uses hidden Markov model for recognizing words directing the movement, such as 'Mae', 'Migi', 'Hidari', 'Tomare', etc. pronounced in Japanese. By using some criterion for the recognition results, we have conducted experiments to validate the human-robot interface using voice commands as well as gesture commands to control the wheel chair properly.