TouchScale: 500 Hours of Human Vision and Touch for Visual-Tactile Learning
Overview
Researchers from Texas A&M, Google DeepMind, Carnegie Mellon University, Stanford University, NVIDIA, and others released TouchScale, a unified dataset comprising approximately 500 hours of human vision and touch data collected using consistent hardware. The study demonstrates that training models on this dataset improves cross-sensor contact prediction accuracy and increases robotic manipulation success rates by an average of 35 percentage points across four tasks, including sorting objects and removing bottle caps, without requiring human action labels or motion redirection.
Generated from attributed reports · Updated 1 days ago
Event evidence and corrections
0 attributed source owners. Ownership does not establish independent confirmation. Quantities are reported separately and are never added together.
Reported quantity · Reported duration: 500 hours · Basis not reported · Differing source assertions
“研究团队构建了总时长500小时的人类视觉-触觉数据集，并从跨传感器触觉预测、视觉表征学习和真实机器人操作三个方向开展验证。”
Exact source · revision 2Source owner not reported
Reported quantity · Robot trajectories: 87000 trajectories · trajectory count
“最终，TouchScale积累了约8.7万条轨迹、涉及约2000种任务描述和1500余件物体。”
Exact source · revision 2Source owner not reported
Reported quantity · Reported duration: 16.2 hours · Basis not reported · Differing source assertions
“研究团队分别使用EgoTouch的16.2小时训练集和TouchScale约16小时子集训练相同的TouchAnything模型，再在独立的EgoTactile数据集上进行零样本评测。”
Exact source · revision 2Source owner not reported
Reported quantity · Reported duration: 16 hours · Basis not reported · Differing source assertions
“研究团队分别使用EgoTouch的16.2小时训练集和TouchScale约16小时子集训练相同的TouchAnything模型，再在独立的EgoTactile数据集上进行零样本评测。”
Exact source · revision 2Source owner not reported
Reported quantity · Reported accuracy: 24.45 percent · Basis not reported · Differing source assertions
“结果显示，TouchScale在三个基准、两种评估设置下均取得了对照组中的最高准确率，平均线性探测准确率达到24.45%，平均微调准确率达到49.62%。”
Exact source · revision 2Source owner not reported
Reported quantity · Reported accuracy: 49.62 percent · Basis not reported · Differing source assertions
“结果显示，TouchScale在三个基准、两种评估设置下均取得了对照组中的最高准确率，平均线性探测准确率达到24.45%，平均微调准确率达到49.62%。”
Exact source · revision 2Source owner not reported
Report timeline
Follow attributed reports and material updates.
- Chinese robotics — Dataset and collection discoverySignalTouchScale: 500 Hours of Human Vision and Touch for Visual-Tactile Learning
Researchers from Texas A&M, Google DeepMind, Carnegie Mellon University, Stanford University, NVIDIA, and others released TouchScale, a unified dataset comprising approximately 500 hours of human vision and touch data collected using consistent hardware. The study demonstrates that training models on this dataset improves cross-sensor contact prediction accuracy and increases robotic manipulation success rates by an average of 35 percentage points across four tasks, including sorting objects and removing bottle caps, without requiring human action labels or motion redirection.
Event coverage history
Current attention 0·Peak within the comparable range 3（2026-11-10 12:00 UTC）·Change within the comparable range over 24 hours –
The trend compares participants observed continuously; its scope may be narrower than current attention totals. Hover or click to explore hourly observations, or use the left and right arrow keys.
Timezone · UTC
Article dates follow your selected timezone. Briefing editions use Hong Kong time (UTC+8).