权益分类	功能权益	普通用户	{{item.name}}会员
{{category.name}}	{{benefitItem.name}}

Environment Understanding using Robot Body

使用机器人本体了解环境

基本信息

批准号：
12308016
负责人：
WADA Toshikazu
金额：
$ 24.36万
依托单位：
Wakayama University (2002)Kyoto University (2000-2001)
依托单位国家：
日本
项目类别：
Grant-in-Aid for Scientific Research (A)
财政年份：
2000
资助国家：
日本
起止时间：
2000 至 2002
项目状态：
已结题

来源：
https://kaken.nii.ac.jp/grant/KAKENHI-PROJECT-12308016/
关键词：
Color Target Detection Nearest Neighbor Classifier Camera Calibration Homography from Circle Environmental Map Generation Discriminating Obstacles and Occluding Objects Correlation between intended action and observed action Behavior Learnin

项目摘要

In this research, we investigated the map generation problem for mobile robots. For the map generation, we made the closed loop consists of robot action and visual feedback verifying its movement. This loop is affected by the interaction between the robot body and its surrounding environment. Based on this idea, we can generate the environmental map, I.e., spatial description of the objects interacting with the robot body. This interaction is detected by finding the inconsistency between the intended action and observed movement. For this research, we first built up specially designed RC cars and its controller connected to the host PC, which also processes the images observed by external vision sensor. Next, in order to tracked the LED attached to the RC cars by the external vision sensor, we proposed a robust color target detection method and developed a prototype system. This system is already used in other institutes. For estimating real motion of the robot on the floor from the observed images, we proposed robot-body guided camera calibration method. This method automatically estimates the camera parameters from the elliptical trajectory on the image plane corresponding to the circular robot motion drawn by keeping a constant steering angle and low speed. This method is especially effective in the case of wide observing area. Based on the calibrated result, real robot movement can be estimated. Using these method, we developed a map generation system, by detecting the inconsistency between intended action and observed movement.

在这项研究中，我们研究了移动机器人的地图生成问题。对于地图生成，我们制作了由机器人动作和验证其运动的视觉反馈组成的闭环。该循环受到机器人本体与其周围环境之间相互作用的影响。基于这个想法，我们可以生成环境地图，即与机器人身体交互的物体的空间描述。通过发现预期动作和观察到的运动之间的不一致来检测这种相互作用。在这项研究中，我们首先构建了专门设计的遥控汽车及其连接到主机 PC 的控制器，主机 PC 还处理外部视觉传感器观察到的图像。接下来，为了通过外部视觉传感器跟踪遥控汽车上的 LED，我们提出了一种鲁棒的颜色目标检测方法并开发了一个原型系统。该系统已在其他机构使用。为了根据观察到的图像估计机器人在地板上的真实运动，我们提出了机器人主体引导相机校准方法。该方法根据与保持恒定转向角和低速绘制的圆形机器人运动相对应的图像平面上的椭圆轨迹自动估计相机参数。该方法在观测区域较宽的情况下尤其有效。根据校准结果，可以估计真实的机器人运动。使用这些方法，我们开发了一个地图生成系统，通过检测预期动作和观察到的运动之间的不一致。