Memories Upload 给第一人称视频按手部动作打时间戳标签
机器人要学动手，靠的是大量真人手部操作视频。这家公司雇人录第一人称画面，按左右手动作逐帧打标签，做成能按动作搜索的训练数据卖给机器人实验室。做具身智能或模仿学习的团队可以直接关注这种数据源。
Memories Upload 推出一个视频检索方案：给头戴摄像头录下的第一人称片段，为每只手逐时刻生成时间戳标注，最终输出 JSON 文件供训练流程使用。这样庞大的视频库不再只能按关键词搜索，而是能按"左手在拧螺丝、右手扶住零件"这类具体动作来查询。
它的底层逻辑是机器人学习路径：机器人通过观察人类的手部操作来学技能。为此 Memories Upload 付费雇佣全球各地的人录制自己工作中的手部画面，再对素材做清洗、打标签，卖给正在训练机器人的实验室。
机器人数据领域长期缺高质量操作类视频，这种"按动作索引"的数据产品把第一人称视频从原始素材变成了可查询的训练数据集，对做 imitation learning 的实验室是直接的供给来源。
Searching video by what the hands are doing is one of the most useful ideas in robotics data right now.
from a camera on a worker’s head to a JSON file in a training run.
Memories Upload, from https://t.co/yaP1XdpqxR, gives every moment of first person footage a time-stamped caption for each hand, so the archive can be queried by action.
Robots learn hand skills by watching humans, so https://t.co/yaP1XdpqxR pays people worldwide to record their hands at work, then cleans, labels and sells that footage to labs building robots.
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. Key entities and technical topics under continuous tracking include 机器人, 第一人称视频, imitation learning, 数据标注.
Source attribution: curated from rohanpaul_ai (@rohanpaul_ai).
