Name: Nvidia 面向仿人機器人的物理人工智能訓練模型 (Nvidia's Physical AI Training Models for Humanoid Robots)
Uploaded: 2025-03-26T14:27:07.000Z
Duration: 3 min 43 s

基礎被動語態

Where the company is showing off a suite of technologies that are designed to help robots learn how to be robots.

在那裡，該公司展示了一套旨在幫助機器人學習如何成為機器人的技術。

One often repeated phrase at this year's Nvidia GTC was physical AI.

在今年的 Nvidia GTC 上，一個經常被重複提及的詞是物理人工智能。

The kind of AI that allows robots to operate effectively in the physical world.

這種人工智能能讓機器人在物理世界中有效運作。

To help get there, Nvidia is offering its Isaac Groot N1.

為了幫助實現這一目標，Nvidia 推出了 Isaac Groot N1。

An open source foundation model that is pre-trained with some of the basics and is broken down into a fast system for taking action and a slow system for planning action.

這是一個開源的基礎模型，預先訓練了一些基礎知識，並分為用於採取行動的快速系統和用於計劃行動的慢速系統。

The next step is teaching the robot new tasks, something that requires a large and diverse data set, especially if the robot will be required to do the learned task in a variety of different conditions.

下一步是教機器人執行新任務，這需要大量不同的數據集，尤其是當機器人需要在各種不同條件下執行所學任務時。

To learn how Nvidia is helping robots get there, we spoke with Akhil Docca, senior product manager for Nvidia's Omniverse.

為了瞭解 Nvidia 如何幫助機器人實現這一目標，我們採訪了 Nvidia Omniverse 高級產品經理 Akhil Docca。

What you're seeing here is essentially the policy training where it's actually learning how to drop something in there.

你在這裡看到的基本上是政策培訓，它實際上是在學習如何在這裡投放東西。

This was actually trained using Teleam, so this particular version of what you see was actually done via Teleam.

這實際上是通過 Teleam 訓練出來的，所以你看到的這個版本實際上是通過 Teleam 完成的。

Where somebody donned on a headset, we do some human demonstration and we scale that using synthetic data.

有人戴上耳機後，我們會進行一些人體演示，然後使用合成數據進行擴展。

The gathering of that synthetic data can be achieved through Nvidia's Omniverse with Cosmos.

通過 Nvidia 的 Omniverse 和 Cosmos，可以收集合成數據。

And Akhil helped me understand what each of those things has to offer.

阿基爾幫助我瞭解了這些東西各自的優點。

Omniverse is platform that brings together data from different sources and allows you to build something like a digital twin, which is physically accurate.

Omniverse 是一個將不同來源的數據彙集在一起的平臺，可以讓您建立一個類似於數字孿生的東西，它在物理上是精確的。

The data that comes out from Omniverse can go through Cosmos.

從 Omniverse 出來的數據可以通過 Cosmos。

Cosmos is really important because it allows not only the augmentation of data for photo realism to reduce the simulation real gap, but it also allows you to create an exponentially large data.

宇宙真的很重要，因為它不僅可以增強數據的照片逼真度，減少模擬的真實差距，還可以創建指數級的龐大數據。

So I can go from maybe a few thousand images to a million images and more, and really that data diversity is important because then it helps you the model in a better way and helps generalize it.

是以，我可以從幾千張圖片到一百萬張圖片，甚至更多，數據的多樣性確實很重要，因為它可以幫助你以更好的方式建立模型，並幫助推廣它。

Going from training around specific tasks to training that results in a more general set of capabilities or behaviors is one big challenge that the whole humanoid robotics industry seems to be working on right now.

從圍繞特定任務進行訓練，到訓練出一套更通用的能力或行為，這是整個仿人機器人行業目前面臨的一大挑戰。

To make that synthetic data as applicable to the real world as possible, requires a detailed representation of the laws of physics.

要使合成數據儘可能適用於現實世界，就需要對物理定律進行詳細的表述。

Today, we're announcing something really, really special, it is a partnership of three companies, Deep Mind, Disney Research and Nvidia, and we call it Newton.

今天，我們要宣佈一件非常非常特別的事情，它是 Deep Mind、迪斯尼研究院和 Nvidia 三家公司的合作項目，我們稱之為 "牛頓"。

Wong brought out a remote controlled BDX droid to celebrate the new physics engine and announced that like Isaac Groot N1, Newton would also be open source.

Wong 帶來了一個遙控 BDX 機器人來慶祝新的物理引擎，並宣佈牛頓也將像 Isaac Groot N1 一樣開源。

That's why Newton is open source because we want to be able to not only give the developers this amazing, you know, work that we're doing with Google Deep Mind and Disney, but also they can contribute back to it.

這也是牛頓開源的原因，因為我們希望不僅能為開發者提供我們與谷歌 Deep Mind 和迪士尼合作的這項令人驚歎的工作，而且開發者也能為這項工作做出貢獻。

The backdrop of much of this work on humanoid robots is a looming anxiety about labor shortages.

仿人機器人工作的主要背景是對勞動力短缺的焦慮。

We know very clearly that the world is has severe shortage of of human laborers, human workers, by the end of this decade, the world is going to be at least 50 million workers short.

我們清楚地知道，世界上的勞動力、工人嚴重短缺，到本十年末，世界上將至少缺少 5000 萬工人。

I reached out to Nvidia who told me the predicted shortage of 50 million workers by the end of the decade was based on a combination of estimates from various sources across various industries and locations.

我聯繫了 Nvidia 公司，他們告訴我，到本世紀末預計將短缺 5000 萬名工人，這是根據不同行業和地區的各種來源的估算綜合得出的。

A representative from Nvidia also clarified the cost model, saying quote, it's more like subscribing to full self-driving services for an autonomous vehicle.

Nvidia 的一位代表還澄清了成本模式，他說，這更像是為自動駕駛汽車訂購全面的自動駕駛服務。

The customer buys the car and then would pay for the services it wants on top.

客戶買了車，然後再支付所需的服務費。

字幕列表影片播放

Nvidia 面向仿人機器人的物理人工智能訓練模型 (Nvidia's Physical AI Training Models for Humanoid Robots)

subscribe

anxiety

specific

essentially