Joyce Chen
    • Create new note
    • Create a note from template
      • Sharing URL Link copied
      • /edit
      • View mode
        • Edit mode
        • View mode
        • Book mode
        • Slide mode
        Edit mode View mode Book mode Slide mode
      • Customize slides
      • Note Permission
      • Read
        • Only me
        • Signed-in users
        • Everyone
        Only me Signed-in users Everyone
      • Write
        • Only me
        • Signed-in users
        • Everyone
        Only me Signed-in users Everyone
    • Invite by email
      Invitee

      This note has no invitees

    • Publish Note

      Share your work with the world Congratulations! 🎉 Your note is out in the world Publish Note No publishing access yet

      Your note will be visible on your profile and discoverable by anyone.
      Your note is now live.
      This note is visible on your profile and discoverable online.
      Everyone on the web can find and read all notes of this public team.

      Your account was recently created. Publishing will be available soon, allowing you to share notes on your public page and in search results.

      Your team account was recently created. Publishing will be available soon, allowing you to share notes on your public page and in search results.

      Explore these features while you wait
      Complete general settings
      Bookmark and like published notes
      Write a few more notes
      Complete general settings
      Write a few more notes
      See published notes
      Unpublish note
      Please check the box to agree to the Community Guidelines.
      View profile
    • Commenting
      Permission
      Disabled Forbidden Owners Signed-in users Everyone
    • Enable
    • Permission
      • Forbidden
      • Owners
      • Signed-in users
      • Everyone
    • Suggest edit
      Permission
      Disabled Forbidden Owners Signed-in users Everyone
    • Enable
    • Permission
      • Forbidden
      • Owners
      • Signed-in users
    • Emoji Reply
    • Enable
    • Versions and GitHub Sync
    • Note settings
    • Note Insights New
    • Make a copy
    • Transfer ownership
    • Delete this note
    • Save as template
    • Insert from template
    • Import from
      • Dropbox
      • Google Drive
      • Gist
      • Clipboard
    • Export to
      • Dropbox
      • Google Drive
      • Gist
    • Download
      • Markdown
      • HTML
      • Raw HTML
Menu Note settings Note Insights Versions and GitHub Sync Sharing URL Create Help
Create Create new note Create a note from template
Menu
Options
Make a copy Transfer ownership Delete this note
Import from
Dropbox Google Drive Gist Clipboard
Export to
Dropbox Google Drive Gist
Download
Markdown HTML Raw HTML
Back
Sharing URL Link copied
/edit
View mode
  • Edit mode
  • View mode
  • Book mode
  • Slide mode
Edit mode View mode Book mode Slide mode
Customize slides
Note Permission
Read
Only me
  • Only me
  • Signed-in users
  • Everyone
Only me Signed-in users Everyone
Write
Only me
  • Only me
  • Signed-in users
  • Everyone
Only me Signed-in users Everyone
  • Invite by email
    Invitee

    This note has no invitees

  • Publish Note

    Share your work with the world Congratulations! 🎉 Your note is out in the world Publish Note No publishing access yet

    Your note will be visible on your profile and discoverable by anyone.
    Your note is now live.
    This note is visible on your profile and discoverable online.
    Everyone on the web can find and read all notes of this public team.

    Your account was recently created. Publishing will be available soon, allowing you to share notes on your public page and in search results.

    Your team account was recently created. Publishing will be available soon, allowing you to share notes on your public page and in search results.

    Explore these features while you wait
    Complete general settings
    Bookmark and like published notes
    Write a few more notes
    Complete general settings
    Write a few more notes
    See published notes
    Unpublish note
    Please check the box to agree to the Community Guidelines.
    View profile
    Engagement control
    Commenting
    Permission
    Disabled Forbidden Owners Signed-in users Everyone
    Enable
    Permission
    • Forbidden
    • Owners
    • Signed-in users
    • Everyone
    Suggest edit
    Permission
    Disabled Forbidden Owners Signed-in users Everyone
    Enable
    Permission
    • Forbidden
    • Owners
    • Signed-in users
    Emoji Reply
    Enable
    Import from Dropbox Google Drive Gist Clipboard
       Owned this note    Owned this note      
    Published Linked with GitHub
    • Any changes
      Be notified of any changes
    • Mention me
      Be notified of mention me
    • Unsubscribe
    # How Open World Models are Powering the Next Breakthroughs in Physical AI ## NVIDIA GTC 2026 產業技術研究報告 --- **Session ID**:S81667 **講者**:NVIDIA VP Generative AI Research **NVIDIA Technology**:Cosmos **主題定位**:Physical AI / World Foundation Models / Sim-to-Real --- ## 一、場次定位與核心命題 Physical AI 的主要瓶頸之一,是「生成足夠逼真(photorealistic)的資料」與「具備穩健推理(reasoning)的世界理解能力」,以支援機器在真實世界的複雜性下仍能可靠感知、理解與適應。NVIDIA Cosmos 世界基礎模型(world foundation models)與其 open models/tools,是用來加速資料產生、訓練客製、與加速部署的核心平台。 本場次將探討開發者如何使用 Cosmos open models and tools 來加速訓練、客製管線、以及更快部署至機器人、自駕車、視覺 AI 代理等場域。同時也會介紹 Cosmos 最新更新如何推進 physics-aware learning、強化 sim-to-real 能力、以及幫助建立下一代智慧適應性 AI 系統。 --- ## 二、背景:Physical AI 為什麼卡在「資料」與「世界理解」 ### 2.1 Physical AI 的資料困境 機器人、AV(自駕/ADAS)、以及視覺型代理(vision AI agents)要在真實環境可靠運作,通常需要: | 資料需求 | 為什麼難 | |----------|----------| | 罕見事件(edge cases)與危險情境 | 真實蒐集昂貴且風險高 | | 多感測器、多視角、跨時間長序列 | 資料量與標註成本爆炸 | | 具備物理一致性的資料 | 動作、碰撞、摩擦、遮擋、慣性等需要正確呈現 | | 跨場域泛化資料 | 不同工廠/道路/室內環境差異大 | Cosmos 的論述是:必須把「世界」本身當成可學習與可生成的對象,用 world foundation models 來支撐合成資料、推理與模擬到真實(sim-to-real)。 ### 2.2 世界模型在 Physical AI 的角色 Cosmos 的研究脈絡把 Physical AI 系統拆成三個部分: | 組成 | 說明 | |------|------| | 機器本體的數位孿生 | 機器人/車輛/設備的虛擬化表示 | | 控制/策略模型(Policy) | 決策與動作生成 | | 世界模型(World Model) | 對環境的理解、預測與推理 | World foundation model 被定位為「可微調成領域世界模型的通用基座」——從通用世界理解能力出發,針對特定場域客製。 --- ## 三、Cosmos 平台組成 Cosmos 被定義為面向 Physical AI 的「developer-first 世界基礎模型平台」,包含以下核心模組: ### 3.1 Cosmos Core Models(世界基礎模型) 一組用於「physics-aware 影片生成」的開放模型套件,包含 diffusion 與 autoregressive transformer 類模型。用來生成世界狀態與影片,或對影片做時空推理。 | 模型特性 | 說明 | |----------|------| | 架構類型 | Diffusion + Autoregressive Transformer | | 核心能力 | Physics-aware 影片生成、世界狀態推理 | | 定位 | 通用世界基座 → 微調為領域專用世界模型 | | 開放性 | Open models,開發者可取用與客製 | ### 3.2 Cosmos Curator(資料處理與策展) 加速資料處理與生成,支援對大量感測資料進行篩選、標註、去重等,降低資料工程成本。 - 資料清理與去重 - 自動標註與屬性化 - 品質篩選與場景分類 - 降低 Physical AI 最大成本來源(資料工程)的人力負擔 ### 3.3 Cosmos Dataset Search(CDS) 對資料集做「即時查詢並取回特定情境(scenarios)」,對於找 edge cases、回放特定事件、建立訓練/驗證切分非常關鍵。 ### 3.4 Guardrails(安全護欄) 在生成/使用世界模型時做安全與治理:輸出限制、內容/場景約束、避免不合理生成擾亂訓練。 ### 3.5 Cosmos Cookbook(實務工作流程) 提供 step-by-step workflows、技術配方與範例,用來重現成功部署並支援客製化。 ### 平台模組總覽 | 模組 | 功能 | 解決的問題 | |------|------|-----------| | Core Models | Physics-aware 影片/世界狀態生成 | 合成資料不足、世界理解能力 | | Curator | 資料清理、去重、標註 | 資料工程成本過高 | | Dataset Search | 情境檢索與子集建立 | Edge case 找不到、測試覆蓋不足 | | Guardrails | 安全護欄與內容約束 | 生成品質控管、治理合規 | | Cookbook | 實作配方與部署範式 | 從模型可用到系統可部署 | --- ## 四、開發者如何用 Cosmos 加速訓練、客製、部署 ### 4.1 資料段:Curator + Dataset Search | 步驟 | 工具 | 目的 | |------|------|------| | 資料清理與去重 | Curator | 降低資料噪音與儲存浪費 | | 自動標註與品質篩選 | Curator | 降低人工標註成本 | | 情境檢索 | Dataset Search | 快速找特定情境,形成訓練/測試集或失效案例集 | 這段的目的是把 Physical AI 最大成本來源(資料工程)變成可規模化的 pipeline。 ### 4.2 模型段:Core Models → 下游客製 | 步驟 | 說明 | |------|------| | 基座選用 | 以 world foundation model 作為通用世界模型基座 | | 領域微調 | 針對你的 domain(工廠、手術室、室內機器人、道路情境)微調成專用世界模型 | | 合成資料生成 | 用客製後的世界模型產生 physics-aware 合成影片/世界狀態 | 這是 Cosmos 論文對「world foundation model → customized world model」的核心定位。 ### 4.3 部署段:工具鏈加速落地 Cookbook 與 docs 的定位是把「模型可用」轉成「系統可部署」:包含資料管線、模型使用方式、護欄與部署範式。 --- ## 五、Physics-Aware Learning 與 Sim-to-Real ### 5.1 為什麼 Physics-Aware 是關鍵 傳統合成資料(如純 3D 渲染)常見的問題: | 問題 | 後果 | |------|------| | 視覺不夠逼真 | Domain gap 導致模型在真實世界退化 | | 物理行為不一致 | 碰撞、慣性、摩擦等不符實際,策略模型學到錯誤行為 | | 場景多樣性不足 | 長尾事件仍然覆蓋不到 | Cosmos 的 physics-aware learning 目標是讓合成世界更貼近真實物理行為,從而讓用合成資料訓練出的模型在真實世界表現更穩健。 ### 5.2 Sim-to-Real 的價值鏈 ``` World Foundation Model(Cosmos Core) │ 領域微調 │ Physics-Aware 合成資料生成 │ 真實 + 合成 混合訓練 │ Sim-to-Real 驗證 │ 真實世界部署 ``` Physics-aware 的程度直接決定 sim-to-real 的 gap 大小。這是本場次最核心的價值主張之一。 --- ## 六、典型落地架構 以下是「用 Cosmos 驅動 Physical AI」的端到端參考架構,可套用到機器人/AV/vision agent 專案: | 步驟 | 內容 | 對應 Cosmos 模組 | |------|------|-----------------| | ① 資料匯入 | 感測資料(影像/多視角影片/雷達等)進資料湖 | — | | ② 資料策展 | 清理、去重、標註/屬性化 | Curator | | ③ 情境檢索 | 建立 scenario 子集(edge cases、長尾事件) | Dataset Search | | ④ 合成資料生成 | 產生 physics-aware 合成影片/世界狀態 | Core Models | | ⑤ 訓練 | 用「真實+合成」混合資料訓練感知/預測/策略模型 | Core Models(微調後) | | ⑥ 驗證 | 以 scenario 子集做 regression、失效追蹤 | Dataset Search + Guardrails | | ⑦ 部署 | 上線到機器人/AV/視覺代理系統 | Cookbook + Guardrails | --- ## 七、應用場域對照 | 應用場域 | 典型場景 | Cosmos 的主要價值 | |----------|----------|-------------------| | 機器人 | 工廠自動化、倉儲 AMR、服務機器人 | 合成場景訓練、physics-aware 避障/導航 | | 自駕車 / ADAS | 道路情境生成、edge case 覆蓋 | 長尾事件合成、sim-to-real 驗證 | | Vision AI Agents | 智慧監控、零售分析、安全巡檢 | 多場景視覺理解、情境推理 | | 醫療 AI | 手術場景模擬、醫療影像生成 | 合成資料補充稀缺臨床資料 | --- ## 八、評估指標與風險點 ### 8.1 建議評估指標 | 面向 | 指標 | |------|------| | 資料面 | 去重率、標註覆蓋率、scenario 覆蓋率(長尾事件比例) | | 模型面 | 真實測試集表現、長尾事件表現、跨 domain 泛化 | | Sim-to-Real 面 | 合成資料比例提升後,真實世界性能是否提升或退化 | | 系統面 | 訓練週期縮短、部署頻率提升、回歸測試時間縮短(MLOps 指標) | ### 8.2 主要風險點 | 風險 | 說明 | 緩解方式 | |------|------|----------| | 合成偏差(Synthetic Bias) | 合成世界與真實分佈偏離,模型學到錯誤捷徑 | 持續監控 real vs synthetic 分佈差異,混合比例調整 | | 物理不一致 | Physics-aware 若不足,策略/控制類模型風險更高 | 以真實資料做 regression 驗證,建立物理一致性 benchmark | | 資料與版本治理 | Scenario 切分、合成版本、模型版本需可追溯 | 完整的 data lineage + model registry | | 安全護欄 | 生成內容、情境約束與輸出可控性 | 導入 Cosmos Guardrails,設定生成邊界與審計 | --- ## 九、決策者應帶走的關鍵結論 | 結論 | 說明 | |------|------| | Physical AI 的瓶頸在資料與世界理解 | 不是模型能力不夠,而是高品質、物理一致的資料太難取得 | | World Foundation Model 是新的基座概念 | 類比 LLM 的基座模型,但目標是「理解並生成世界」 | | Open models 降低導入門檻 | 開發者可直接取用 Cosmos 開放模型,客製成領域專用 | | 資料工程是最大成本 | Curator + Dataset Search 把資料管線規模化是導入的第一步 | | Physics-aware 程度決定 sim-to-real gap | 合成資料的物理一致性直接影響真實部署性能 | | Guardrails 是量產必備 | 安全護欄不是附加功能,是生成式 AI 用於 Physical AI 的前提 | | 先用 Cookbook 跑通再客製 | 從官方配方出發,驗證可行後再做領域微調 | --- ## 十、延伸學習資源 | 主題 | 建議資源 | |------|----------| | GTC26 S81667 場次資訊 | GTC Session Catalog(場次摘要) | | Cosmos 官網 | NVIDIA Cosmos 官方頁(平台模組:Core Models、Curator、Dataset Search) | | Cosmos 官方文件 | NVIDIA Cosmos Docs(平台定義、guardrails、資料管線) | | Cosmos Cookbook | Cosmos Cookbook(step-by-step 實作配方與工作流程) | | Cosmos WFM 技術介紹 | NVIDIA Blog:Cosmos World Foundation Models | | Cosmos 研究論文 | arXiv:Cosmos World Foundation Model Platform | | 先修內容 | GTC25:An Introduction to NVIDIA Cosmos World Foundation Models | | 會後回看 | NVIDIA On-Demand(會後以 S81667 搜尋錄影/投影片) | --- *— 報告完 —*

    Import from clipboard

    Paste your markdown or webpage here...

    Advanced permission required

    Your current role can only read. Ask the system administrator to acquire write and comment permission.

    This team is disabled

    Sorry, this team is disabled. You can't edit this note.

    This note is locked

    Sorry, only owner can edit this note.

    Reach the limit

    Sorry, you've reached the max length this note can be.
    Please reduce the content or divide it to more notes, thank you!

    Import from Gist

    Import from Snippet

    or

    Export to Snippet

    Are you sure?

    Do you really want to delete this note?
    All users will lose their connection.

    Create a note from template

    Create a note from template

    Oops...
    This template has been removed or transferred.
    Upgrade
    All
    • All
    • Team
    No template.

    Create a template

    Upgrade

    Delete template

    Do you really want to delete this template?
    Turn this template into a regular note and keep its content, versions, and comments.

    This page need refresh

    You have an incompatible client version.
    Refresh to update.
    New version available!
    See releases notes here
    Refresh to enjoy new features.
    Your user state has changed.
    Refresh to load new user state.

    Sign in

    Forgot password
    or
    Sign in via Google Sign in via Facebook Sign in via X(Twitter) Sign in via GitHub Sign in via Dropbox Sign in with Wallet
    Wallet ( )
    Connect another wallet

    New to HackMD? Sign up

    By signing in, you agree to our terms of service.

    Help

    • English
    • 中文
    • Français
    • Deutsch
    • 日本語
    • Español
    • Català
    • Ελληνικά
    • Português
    • italiano
    • Türkçe
    • Русский
    • Nederlands
    • hrvatski jezik
    • język polski
    • Українська
    • हिन्दी
    • svenska
    • Esperanto
    • dansk

    Documents

    Help & Tutorial

    How to use Book mode

    Slide Example

    API Docs

    Edit in VSCode

    Install browser extension

    Contacts

    Feedback

    Discord

    Send us email

    Resources

    Releases

    Pricing

    Blog

    Policy

    Terms

    Privacy

    Cheatsheet

    Syntax Example Reference
    # Header Header 基本排版
    - Unordered List
    • Unordered List
    1. Ordered List
    1. Ordered List
    - [ ] Todo List
    • Todo List
    > Blockquote
    Blockquote
    **Bold font** Bold font
    *Italics font* Italics font
    ~~Strikethrough~~ Strikethrough
    19^th^ 19th
    H~2~O H2O
    ++Inserted text++ Inserted text
    ==Marked text== Marked text
    [link text](https:// "title") Link
    ![image alt](https:// "title") Image
    `Code` Code 在筆記中貼入程式碼
    ```javascript
    var i = 0;
    ```
    var i = 0;
    :smile: :smile: Emoji list
    {%youtube youtube_id %} Externals
    $L^aT_eX$ LaTeX
    :::info
    This is a alert area.
    :::

    This is a alert area.

    Versions and GitHub Sync
    Get Full History Access

    • Edit version name
    • Delete

    revision author avatar     named on  

    More Less

    Note content is identical to the latest version.
    Compare
      Choose a version
      No search result
      Version not found
    Sign in to link this note to GitHub
    Learn more
    This note is not linked with GitHub
     

    Feedback

    Submission failed, please try again

    Thanks for your support.

    On a scale of 0-10, how likely is it that you would recommend HackMD to your friends, family or business associates?

    Please give us some advice and help us improve HackMD.

     

    Thanks for your feedback

    Remove version name

    Do you want to remove this version name and description?

    Transfer ownership

    Transfer to
      Warning: is a public team. If you transfer note to this team, everyone on the web can find and read this note.

        Link with GitHub

        Please authorize HackMD on GitHub
        • Please sign in to GitHub and install the HackMD app on your GitHub repo.
        • HackMD links with GitHub through a GitHub App. You can choose which repo to install our App.
        Learn more  Sign in to GitHub

        Push the note to GitHub Push to GitHub Pull a file from GitHub

          Authorize again
         

        Choose which file to push to

        Select repo
        Refresh Authorize more repos
        Select branch
        Select file
        Select branch
        Choose version(s) to push
        • Save a new version and push
        • Choose from existing versions
        Include title and tags
        Available push count

        Pull from GitHub

         
        File from GitHub
        File from HackMD

        GitHub Link Settings

        File linked

        Linked by
        File path
        Last synced branch
        Available push count

        Danger Zone

        Unlink
        You will no longer receive notification when GitHub file changes after unlink.

        Syncing

        Push failed

        Push successfully