资源简介:
---
license: mit
language:
- en
tags:
- code
- robotics
- manipulation
- modulus
- youngs
- youngs modulus
- compliance
- hardness
- tactile
- tactile sensing
- robotic manipulation
- gelsight
- shore
- shape
- material
---
GelSight Young's Modulus Dataset
==============
by [Michael Burgess](https://mburgjr.github.io/)
Dataset of tactile images collected over grasping common objects labelled with the objects' Young's Moduli. All images are collected using GelSight Wedges, both with and without displacement markers. Associated code is available on [GitHub](https://github.com/GelSight-lab/EstimateModulus/tree/main).
Dataset Format
---------------
Each object is grasped a number of times, and each of these grasps is downsampled to 3 frames and shifted to create different data augmentations of the same grasp. Collected data is split into folders according to the following structure...
```
gelsight_youngs_modulus_dataset
└── {object_name}
├── metadata.json
└── grasp={grasp_number}
└── augmentation={augmentation_number}
├── RGB.pkl
├── depth.pkl
├── RGB_markers.pkl
├── depth_markers.pkl
├── forces.pkl
├── widths.pkl
├── elastic_estimate.pkl
└── hertz_estimate.pkl
```
In the OBJECT_NAME folder, metadata for each object is provided in a .json file, including the object's shape, material, Young's modulus, and Shore hardness.
```
metadata = {
'object_name': {object_name}, # [str]
'youngs_modulus': 0.0, # [Pa]
'material': '', # [str]
'shape': '', # [str]
'shore_00_hardness': None, # [/]
'shore_A_hardness': None, # [/]
'shore_D_hardness': None, # [/]
'used_in_training': True # [bool]
}
```
In the AUGMENTATION_NUMBER folder, data is provided in .pkl files. You will find the following named conventions...
- Tactile RGB images (3 images, without markers): `RGB.pkl`
- Tactile depth images (3 images, without markers): `depth.pkl`
- Tactile RGB images (3 images, with markers): `RGB_markers.pkl`
- Tactile depth images (3 images, with markers): `depth_markers.pkl`
- Sampled grasping contact forces: `forces.pkl`
- Sampled gripper widths: `widths.pkl`
- Elastic analytical model estimate: `elastic_estimate.pkl`
- Hertzian analytical model estimate: `hertz_estimate.pkl`
### 许可证:MIT许可证
### 语言:
- 英语
### 标签:
- 代码
- 机器人学
- 操控
- 模量
- 杨氏
- 杨氏模量(Young's Modulus)
- 柔度
- 硬度
- 触觉
- 触觉传感(Tactile Sensing)
- 机器人操控(Robotic Manipulation)
- GelSight
- 邵氏(Shore)
- 形状
- 材料
# GelSight杨氏模量数据集(GelSight Young's Modulus Dataset)
作者:[Michael Burgess](https://mburgjr.github.io/)
本数据集收录了抓取常见物体时采集的触觉图像,所有图像均标注了对应物体的杨氏模量。所有图像均使用GelSight楔形传感器采集,部分采集过程带有位移标记,部分无标记。配套开源代码已发布于[GitHub](https://github.com/GelSight-lab/EstimateModulus/tree/main)。
## 数据集格式
每个物体被抓取多次,每次抓取的图像序列会被下采样至3帧,并通过移位操作生成同一抓取动作的多种数据增强样本。采集得到的数据按以下目录结构组织:
gelsight_youngs_modulus_dataset
└── {object_name} # 物体名称
├── metadata.json # 元数据文件
└── grasp={grasp_number} # 抓取编号
└── augmentation={augmentation_number} # 数据增强编号
├── RGB.pkl
├── depth.pkl
├── RGB_markers.pkl
├── depth_markers.pkl
├── forces.pkl
├── widths.pkl
├── elastic_estimate.pkl
└── hertz_estimate.pkl
在物体名称对应的文件夹中,包含该物体的元数据JSON文件,其中涵盖了物体的形状、材料、杨氏模量以及邵氏硬度等信息。
metadata = {
"object_name": "{object_name}", # 字符串类型,物体名称
"youngs_modulus": 0.0, # 浮点型,单位为帕斯卡(Pa)
"material": "", # 字符串类型,材料名称
"shape": "", # 字符串类型,物体形状
"shore_00_hardness": null, # 邵氏00型硬度值
"shore_A_hardness": null, # 邵氏A型硬度值
"shore_D_hardness": null, # 邵氏D型硬度值
"used_in_training": true # 布尔类型,是否用于训练集
}
在数据增强编号对应的文件夹中,所有数据均以.pkl文件格式存储,各文件的命名规则如下:
- 无位移标记的触觉RGB图像(共3帧):`RGB.pkl`
- 无位移标记的触觉深度图像(共3帧):`depth.pkl`
- 带位移标记的触觉RGB图像(共3帧):`RGB_markers.pkl`
- 带位移标记的触觉深度图像(共3帧):`depth_markers.pkl`
- 采样得到的抓取接触力数据:`forces.pkl`
- 采样得到的夹爪开度数据:`widths.pkl`
- 弹性解析模型的估计结果:`elastic_estimate.pkl`
- 赫兹解析模型的估计结果:`hertz_estimate.pkl`