Physical AI: Robots Lack Touch, Not Intelligence
物理AI：机器人不缺脑子，缺的是手感
This article discusses the development of physical AI through the creation of high-quality datasets for real-world interaction. It outlines the challenges of data scarcity, cost, and the three main approaches to data collection: real-world teleoperation, simulation, and human video recording. The article also highlights the strategic initiative by Jiangsu province to build these datasets and the importance of standardization, data sharing, and ethical considerations in the development of physical AI systems.
Source: Chinese robotics — Dataset and collection discovery · 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
data collection hours
≈500,000 hours
data collection hours
>1,000,000 hours
data collection cost
200,000,000
data collection cost
10–15
data collection hours
22
- code: not_reported
- dataset: not_reported
- weights: not_reported
- paper: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
行业里流传的一组数字：国内真实物理交互场景的合规数据只有约50万小时，而机器人真正商业化落地需要数千万小时。
Open source S15
一条30秒的真机操作数据，采集成本在10到15元；业内测算，要把数据量凑到20万小时，得花2亿元以上。
Open source S19
Meta 那个用视频学习物理规律的模型，一口气喂了超过100万小时，业界认为离终点还远。
Open source S21
可这20万小时连续放起来，不过22年——对于预训练来说，仍然是杯水车薪。
Open source S20
而数千万小时，是三千年往上。
Open source S18
Source：Chinese robotics — Dataset and collection discovery · sohu.com