Embodied AI data services grow amid standardization and privacy gaps
创业邦 科技 reports that companies are emerging to collect, annotate and train embodied-AI datasets, including rental-based household scenarios, dedicated training grounds and crowdsourced collection. It says Zhiyuan Robotics subsidiary 觅蜂科技 had expanded crowdsourced collection to 40 cities, while JD plans to mobilize employees and external participants to gather 5 million hours of video in one year and more than 10 million within two years. However, it cites a China Academy of Information and Communications Technology report saying collection, annotation and robot configurations remain unstandardized, while commercial viability and privacy compliance remain uncertain.
The report indicates that real-world training data is becoming a distinct service layer around embodied AI, but differences in robot configurations and unresolved privacy and business-model issues limit how broadly the data can be reused.
谁能想到,具身智能数据采集这项工作的第一次出圈,竟然是因为一次租房乌龙。
最近有则新闻上了热搜:有人一口气租下了30套房子,而且每套房子只租一周左右,以至于中介都误以为这是什么新型骗局或是有什么不法勾当,一通调查下来,才发现租这么多房子,只是为了搭场景为人形机器人采集训练数据。
这个行业为了采集数据,已经卷到这种地步了吗?雷科技(ID:leitech)发现,这样的故事不是个例。
随着具身智能赛道越发火热,一批专门为机器人提供数据采集、数据标注、训练服务的公司开始涌现。
有人建起数千平方米的机器人训练场,有人把采集设备发给普通人,让大家在家里做家务的同时顺便给机器人采集数据,还有人直接把业务铺到了全国几十个城市。
机器人本体还在努力寻找商业化场景,围绕它们学习技能的数据采集工作,却已经先一步热闹起来了。