TouchScale: 500 Hours of Human Vision and Touch for Visual-Tactile Learning
把人类双手的「手感」移植给机器人：500 小时真实操作数据，任务成功率翻番
TouchScale is a 500-hour human visual-tactile dataset collected with unified sensors and synchronization, showing scaling benefits in both perception and robotic manipulation. The dataset enables cross-sensor generalization and improves robot task success rates from 22.5% to 57.5% through mid-training. The study highlights the potential of large-scale visual-tactile data to enhance embodied AI systems.
Source: Machine Heart — Robotics on WeChat · Read original article ↗
English translation unavailable. Choose the original article to read it.
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
Reported numbers
Reported duration
500 hours
Reported duration
500 hours
Reported success rate
22.5%
Reported success rate
57.5%
- paper: Available Artifact link Open source S20
- code: not_reported
- weights: not_reported
- dataset: Available Artifact link Open source S20
What remains unknown
Not established in the collected evidence: Environment, Control, Data origin.
Reported performance applies to the described task. It does not establish general autonomy or deployment readiness.
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
把人类双手的「手感」移植给机器人：500 小时真实操作数据，任务成功率翻番
Open source S1
TouchScale 包含 500 小时人类双手视觉-触觉同步数据，全部由同一套穿戴式系统采集，同步记录头部 RGB-D、双腕 RGB 和双手全手触觉信号。
Open source S14
无需人类动作标签，也不把人手动作映射到机器人，只用人类视触觉数据做 mid-training，机器人 4 个真机接触任务的平均成功率从 22.5% 升到 57.5%。
Open source S16
题目：TouchScale: 500 Hours of Human Vision and Touch for Visual-Tactile Learning 论文链接： https://arxiv.org/abs/2610.10288 项目主页：https://touch-scale.github.io/ 数据集地址：https://huggingface.co/datasets/2077AIDataFoundation/TouchScale 500 小时触觉数据，难的不只是「采得多」 对 TouchScale 来说，真正的难点不是把小时数简单堆到 500，而是在同一套传感器、同一套同步与采集流程下持续扩展。
Open source S20
Source：Machine Heart — Robotics on WeChat · mp.weixin.qq.com