Huggingface 上线 44.7TB 开源机器人第一人称视频数据集
做机器人训练的朋友可以关注这个,10 万小时级的第一人称作业视频还允许商用,比自己在实验室录数据便宜太多。12,000 多个不同家庭和车间的场景对解决 sim-to-real 泛化问题很有价值,需要深度、夹爪等格式还能找 Hub 定制录制。
Hub 在 Huggingface 发布了一个 44.7TB 的开源机器人训练数据集,包含 10,000 小时人类执行熟练工作的第一人称视频,许可允许商业使用。未来几周还将追加 90,000 小时,总贡献者超过 100,000 人。
该数据集的差异化在于环境多样性:已拍摄 12,227 名操作者,每人在自己的车间或家中作业,因此每个新人都带来新环境,直接针对实验室训练的机器人换到陌生房间就失效的问题。
Hub 背后运营着 860K 名经过验证的操作者网络,覆盖 7,300 多个商业站点。AI 实验室可付费定制录制更丰富的格式,包括立体视觉、深度、手部追踪、夹爪数据、遥操作,以及在真实站点部署机器人采集数据。
An insanely large open source robotics dataset just dropped on Huggingface. 44.7 TB
@hubdotxyz just released 10,000 hours of first-person video of people doing skilled jobs, free even for commercial use. Another 90,000 hours, from more than 100,000 contributors in total, arrive over the coming weeks.
Robots trained in a lab often fail in an unfamiliar room. but this dataset films 12,227 people, each in their own workshop or home, so every new person brings a new environment.
Hub runs a network of 860K verified operators at 7,300+ commercial sites. AI labs pay it to record custom data on request: richer formats (stereo, depth, hand tracking, gripper data), teleoperation, and robots deployed on real sites.
This report is categorized under AITOP's "Products" desk, focusing on US frontier AI labs and Silicon Valley technology leaders. Within our cross-regional analytical framework, it directly maps to AI developer toolchains, coding agents, and engineering workflows and US-China foundation model benchmarks and commercialization comparison. Key entities and technical topics under continuous tracking include Huggingface, Hub.
Source attribution: curated from rohanpaul_ai (@rohanpaul_ai).