Online Acquisition of Close-Range Proximity Sensor Models for Precise Object Grasping and Verification
Online Acquisition of Close-Range Proximity Sensor Models for Precise Object Grasping and Verification
复制标题
DOI：
10.1109/lra.2020.3010440
复制
发表时间：
2020-10-01
影响因子：
5.2
通讯作者：
Inaba, Masayuki
中科院分区：
文献类型：
--
作者：
Hasegawa, Shun;Yamaguchi, Naoya;Inaba, Masayuki
本研究提出一种方法，用于获取近距离近似接近传感器的模型参数的机器人手使用远程距离传感器，而手是抓住一个物体。获取的模型用于生成精确的近距离距离输出。我们在此的目标是获得对对象属性具有很小依赖性并且可以感测宽范围（即，近距离和远距离）。简单的近距离传感器强烈地依赖于对象属性，例如反射率、材料、体积和/或导电性，而远距离传感器不能精确地测量近距离。为了实现我们的目标，我们融合了近距离和远程传感器。简单的融合在近距离仍然依赖于对象。因此，我们在近距离传感器模型中使用两种传感器类型重叠处的远距离传感器的距离输出来获得对象相关参数。通过真实的机器人实验，我们评估了在近距离产生的距离输出的精度，并发现它是有用的柔顺物体的精确抓取。我们还证实，所获得的对象相关的参数可以验证超薄对象的抓取。
This study presents an approach for acquiring model parameters of close-range approximate proximity sensors on a robot hand using long-range distance sensors while that hand is grasping an object. The acquired models are used to generate a precise close-range distance output. We aim herein to acquire proximity sensors that have little dependence on object properties and that can sense a wide range (i.e., both close and long ranges). Simple close-range sensors strongly depend on object properties such as reflectance, material, volume, and/or conductivity, whereas long-range sensors cannot precisely measure the close range. To accomplish our goal, we fused close- and long-range sensors. Simple fusion remains object dependent at the close range. Hence, we acquired an object-dependent parameter in the close-range sensor model using the distance output of the long-range sensor at the overlap of the two sensor types. Through real robot experiments, we evaluated the precision of the generated distance output at the close range and found it useful to the precise grasping of compliant objects. We also confirmed that the acquired object-dependent parameter can verify ultra-thin object grasping.