SSVTP

SSVTP数据集由加州大学伯克利分校AUTOLab联合德累斯顿工业大学LASR实验室等机构创建，旨在通过自监督学习实现视觉与触觉数据的多模态融合。该数据集包含4500对空间对齐的视觉与触觉图像，采集自10种不同柔性表面环境，涵盖服装特征如纽扣、拉链、边缘等。数据采集借助定制机械臂与传感器完成，确保图像对齐且具有相似视野范围。数据集用于训练视觉触觉共享潜空间，支持无需微调的下游任务，如异常检测、特征搜索、边缘跟踪等。该数据集为机器人在柔性物体操作领域的多模态感知与控制提供支持，助力机器人自主学习与任务执行。

The SSVTP dataset was created in collaboration by AUTOLab at the University of California, Berkeley, the LASR Lab at Technische Universität Dresden, and other research institutions, with the objective of achieving multimodal fusion of visual and tactile data through self-supervised learning. This dataset contains 4500 spatially aligned pairs of visual and tactile images, collected from 10 different flexible surface environments, and covers clothing-related features such as buttons, zippers, and edges. Data collection was conducted using custom-built robotic arms and sensors, ensuring image alignment and consistent field-of-view ranges across all samples. The dataset is designed for training visual-tactile shared latent spaces, and supports downstream tasks that require no fine-tuning, including anomaly detection, feature search, edge tracking, and other related tasks. This dataset supports multimodal perception and control for robots in the domain of flexible object manipulation, and facilitates autonomous learning and task execution for robotic systems.

搜集汇总

背景与挑战

背景概述

SSVTP数据集由加州大学伯克利分校AUTOLab联合德累斯顿工业大学LASR实验室等知名机构精心打造，是机器人多模态感知领域的创新资源。该数据集包含4500对空间对齐的视觉与触觉图像，涵盖10种不同柔性表面环境，细致记录了服装特征如纽扣、拉链、边缘等关键细节。数据采集借助先进的定制机械臂与高精度传感器完成，确保图像对齐且视野范围高度一致，为多模态融合研究提供了坚实基础。