機器人在乾淨的實驗室學會拿香蕉,到了真實廚房卻可能不知道怎麼動手。
- Hub 公開了 10,000 小時人類實際工作的資料,來自超過 12,000 人,場景包括水電修繕、焊接、微型焊接和修理手錶,環境也遍佈各處。
- 這些工作帶有日常生活中的混亂和變化,或許能讓機器人接觸比實驗室更接近真實世界的情況。
- Hub 還表示,另外 90,000 小時資料將陸續提供。
AI 伴隨問答
針對這篇報導有疑問?可直接點選快捷問題或向 AI 追問背景脈絡與技術細節:
Robotics has spent years teaching million-dollar machines to pick up the same banana in a spotless laboratory.
Then the poor things encounter a real kitchen and suddenly forget how hands work.
This is why what Hub just released is so important.
10,000 hours of actual humans doing actual work. Over 12,000 people across thousands of environments. Plumbing, welding, microsoldering, watch repair.
All the wonderful, messy, unpredictable things that make the physical world so difficult.
And they’re giving it away. Another 90,000 hours on the way.
Turns out the secret to teaching robots how to live in our world might be showing them just how gloriously chaotic it actually is.
Who would have thought 🤷🏻♂️