Perceptron AI 发布 EgoHome 基准，GPT-6 Astra 在手部动作理解测试中居首
机器人靠人类视频学动作，视频标签全靠 AI 自动打，标签错了机器人就学废了。Perceptron AI 搞了个 EgoHome 基准专门测这个，GPT-6 Astra 拿了第一。做具身智能或数据标注 pipeline 的团队应该看看。
Perceptron AI 推出名为 EgoHome 的新基准，专门测试 AI 能否识别 chore 视频中人类手部的具体动作，GPT-6 Astra 在该基准上排名第一。
EgoHome 聚焦的不是“有人在洗碗”这类场景级描述，而是“左手拿起海绵、右手拧水龙头”级别的细粒度动作标注，这正是机器人向人类视频学习所需要的能力。目前机器人训练大量依赖人类视频，并用 AI 自动给视频打动作标签；
如果标签出错，机器人就会学到错误的行为。这一基准为评估标注模型的可靠性提供了量化工具。
GPT-6 Astra came out on top in EgoHome, a new benchmark from Perceptron AI, that tests whether AI can follow what human hands do in chore videos, the skill robots need to learn from people.
AI can tell you someone is washing dishes, but not what their hands are doing, and robots need the second part.
This kind of benchmarks are so important because Robotics training now happen on large amounts of human video, and AI is used to label those videos automatically (“left hand picks up sponge, right hand turns tap”). If those labels are wrong, the robots learn wrong.
https://t.co/t5gL9ZYYmU
This report is categorized under AITOP's "Models" desk, focusing on US frontier AI labs and Silicon Valley technology leaders. Within our cross-regional analytical framework, it directly maps to US-China foundation model benchmarks and commercialization comparison. Key entities and technical topics under continuous tracking include GPT-6 Astra, EgoHome, 机器人, 基准测试.
Source attribution: curated from rohanpaul_ai (@rohanpaul_ai). Cross-source corroboration: 11 independent sources have covered this development (including Runway ML, Epoch AI, berryxia), confirming industry-wide signal strength.
