查看購物車
  • English
  • Japanese
  • Korean
封面
市場調查報告書
商品編碼
2064026

嵌入式人工智慧（人形機器人）用主控SoC研究報告（2026）

Embodied AI (Humanoid Robot) Main Control SoC Research Report, 2026
出版日期: | 出版商: ResearchInChina | 英文 400 Pages | 商品交期: 最快1-2個工作天內

價格
簡介目錄

人工智慧SoC實現調查：晶片供應商正在從「單一SoC供應商」轉型為「全端晶片平台供應商」。

晶片技術的進步為快速發展的嵌入式人工智慧（EAI）產業提供了決定性的推動力。不同的機器人對晶片的選擇要求各不相同，必須避免因選擇不當而導致的問題，例如運算能力過強或效能不足。此外，EAI產業的發展也仰賴大規模建模技術的突破。機器人的智慧水平已顯著提升，使其能夠自主決策並執行複雜任務。

隨著企業應用整合 (EAI) 市場的擴張和晶片效能要求的提高，晶片供應商正在部署全端解決方案。

機器人晶片市場正經歷快速成長。預計2025年，全球通用EAI機器人的出貨量將達到13,000台，並在2026年突破50,000台。目前，各大晶片廠商正在推出以EAI為導向的SoC晶片，例如NVIDIA Jetson系列和高通IQ10系列。同時，為了滿足客戶對快速應用部署和模型開發的需求，各公司也紛紛推出機器人開發平台，例如NVIDIA Isaac開放原始碼平台、第二代瑞芯微RKNN神經網路模型轉換和最佳化工具以及黑芝麻SmartX多維智慧運算平台等。

目前，企業應用人工智慧（EAI）的系統單晶片（SoC）正以以下方式發展：

趨勢一：對晶片運算能力的需求顯著增加。

NVIDIA 的新款 Jetson T5000 採用 Blackwell GPU 架構，AI 運算效能高達 2070 FP4 TFLOPS，是上一代 Jetson Orin 的 7.5 倍。 Horizo​​n Robotics（D-Robotics）的 RDK S100P 將 CPU、BPU 和 MCU 整合在單一晶片上，提供 120 TOPS 的運算效能。隨著演算法日益複雜，機器人所需的運算能力正從目前的 200-500 TOPS 逐步提升至 500-1000 TOPS。尤其值得注意的是，業界正從單純追求運算能力轉向優先考慮效率。透過演算法最佳化，效率已成為一項核心指標。

趨勢二：晶片供應商正在向先進製程發展。

主流晶片廠商正朝著更先進的製程製程方向發展。 NVIDIA 的 Jetson AGX Thor 採用 4nm 製程，Intel 的 Core Ultra 系列 3 採用 Intel 18A 製程，Rockchip 的 RK3588 採用 8nm 製程，聯發科最新的 Genio Pro 採用 3nm 製程，所有這些都顯著提升了晶片性能。

趨勢 3：EAI 晶片供應商正在從「單一 SoC 供應商」轉型為「全端晶片平台供應商」。

以SemiDrive為例，除了EAI大腦SoC之外，他們還銷售智慧控制小腦SoC和高效能MCU，建構了涵蓋「大腦-小腦-身體-關節」完整架構的全端EAI解決方案。該公司的產品線十分豐富，從處理高階認知和決策的主控大腦SoC，到處理運動調整和即時控制的智慧控制小腦晶片，再到用於雷射雷達/機器視覺、運動中心、敏捷手和關節模組的E3-R系列MCU，涵蓋了完整的晶片鏈。

其中，用於智慧控制的「D9-Max」小腦晶片和用於機器人關節模組的「E311x-R」MCU已進入量產階段。與多家大型機器人公司建立了緊密的合作關係，成功將車規級高性能和高可靠性技術引入機器人領域。

D9-Max 採用專為小腦應用最佳化的架構。基於硬體隔離和硬體虛擬化技術，它整合了一個 8 核心 2.0GHz Cortex-A55 CPU叢集、一個 4 核心 2.0GHz Cortex-A55 CPU叢集、三對 800MHz Cortex-R5F 雙核心同步處理器，以及 8TOPS NPU 和 GPU 等運算單元。運動控制系統、人機介面 (HMI) 和 EtherCAT 主站這三大核心功能可部署在單一晶片上，將以往需要三個晶片才能實現的功能整合到了一個晶片中。

高效能MCU（E3-R系列）代表了關節控制領域的重大進步，滿足高功能安全性和網路安全要求，並提供一站式解決方案。作為關節模組的主控晶片，E311x-R擁有極高的即時效能和極為穩定的運算能力。它採用雙R5F內核，主時脈頻率高達400MHz。在實際研發中，雙核心將馬達控制和通訊處理分開，並透過分配專用核心來提升效能。

在EAI大腦SoC方面，SemiDrive利用其在汽車領域設備端大規模模型處理方面的專業知識，開發了下一代機器人大腦晶片「R1」。它採用ARM V9.2架構CPU和新型高效能NPU，支援在低功耗下對MLLM/VLA等嵌入式端對端模型進行設備端部署。

趨勢 4：晶片供應商正在部署全端、自主研發的工具鏈。

瑞芯微發布了第二代神經網路模型轉換和最佳化工具RKNN-Toolkit2。該工具旨在連接主流深度學習框架和瑞芯微的NPU（神經處理單元）硬體平台，使開發者能夠有效地將訓練好的AI模型部署到嵌入式設備。黑芝麻科技基於華山A2000構建了方便用戶使用的「山海AI」工具鏈，涵蓋從模型最佳化到設備端部署的整個流程，為開發者提供高效的模型開發和部署系統。 SemiDrive提供全面的軟硬體開發套件，例如「D9-Max」應用開發套件，使客戶和獨立開發者能夠快速部署應用程式並在裝置上進行開發。

OEM廠商選擇晶片和演算法來實現機器人功能。

EAI（企業應用智慧）水平本質上是演算法和晶片共同演化的結果。兩者相互依存，相互促進，形成封閉回路型，而這正是機器人智慧系統的核心。

例如，AgiBot Lingxi X2 的基礎運算板採用了兩顆瑞芯微 RK3588 晶片，取代了上一代產品使用的 Jetson Xavier，在成本和性能方面均有所提升。 RK3588 的 6 TOPS NPU 在運動控制和感知融合場景下表現出色，同時功耗降低了 7W。高效能闆卡則採用了 NVIDIA Jetson Orin NX，總合AI 運算能力高達 169 TOPS。

在演算法方面，凌曦X2的「大腦」搭載了AgiBot自主研發的大規模模型「Genie Operator-1 (GO-1)」。它採用視覺-語言-潛在動作（ViLLA）架構，該架構由多模態大規模模型（VLM）和專家混合模型（MoE）組成，使凌曦X2擁有卓越的學習能力、透過小樣本學習實現的快速泛化能力以及持續進化能力。凌曦X2的「小腦」則採用了Xyber-Edge控制器，負責協調機器人的運動並做出決策。該控制器採用 144 核異構計算架構，動態地將推理任務分配給 NPU叢集，將控制命令分配給 FPGA，將傳統的 12 層控制架構的運動規劃過程壓縮為 3 層隱式規劃結構，實現了 450Hz 的即時封閉回路型控制——顯著超過了 Tesla Optimus 的 280Hz封閉回路型頻率。

AgiBot推出了三條產品線——「元正」、「靈曦」和「吉尼」——每條產品線都針對工業製造、商業服務和數據研究場景進行了差異化和互補性的發展，並正朝著大規模生產和商業部署的方向邁進。

目錄

第1章：企業應用整合（EAI）市場及應用場景

  • 企業應用整合（EAI）的基本概念和術語
  • 企業應用整合（EAI）的基本概念
  • EAI術語
  • EAI市場展望
  • 企業應用整合（EAI）的演變
  • EAI產業的現狀
  • 企業應用整合（EAI）應用場景的演變
  • 企業應用整合（EAI）市場趨勢
  • 中國EAI市場規模
  • 全球人形機器人出貨量
  • EAI的應用前景
  • 預期申請人概況
  • 按應用場景分類的EAI市場結構
  • 社區與家庭場景：居家服務（1）
  • 社區與家庭場景：居家服務（2）
  • 社區/家庭場景：醫療/護理場景（1）
  • 社區/家庭場景：醫療/護理場景（2）
  • 智慧製造場景：工廠生產
  • 智慧製造場景：人形機器人輔助工廠實現全天候運作
  • 智慧製造場景：優必選沃克S2團隊特定智慧工廠專案中的協作工作
  • 智慧製造場景：農業生產
  • 商業服務場景：KEENON 機器人
  • 商業服務範例：美團「小峰」
  • 高風險救援場景：DEEP Robotics LYNX M20 輪式機器人
  • 高風險救援場景：iFreecomm公司的「靈木」緊急救援四足機器人和導盲犬
  • EAI供應商競爭格局概述
  • 中國排名前 50 名的 EAI 供應商
  • 十大海外EAI供應商
  • 2025年全球十大人形機器人出貨量（基於主流統計數據）
  • 一家典型的EAI公司的收入
  • 領先的EAI公司的技術路線

第2章：企業應用整合（EAI）的軟體與硬體系統結構

  • EAI硬體架構
  • EAI：硬體系統簡介
  • EAI硬體清單
  • EAI晶片列表
  • SemiDrive：面向機器人領域的全端晶片解決方案
  • GigaDevice：面向機器人的全端晶片解決方案
  • 英飛凌：人形機器人功能模組解決方案
  • 英飛凌：人形機器人產品佈局
    • EAI硬體系統：運算能力與硬體控制系統
    • EAI硬體系統：運算能力與硬體控制系統
    • EAI硬體系統：Cerebro系統的配置
    • EAI硬體系統：「大腦」系統－主控SoC的應用
    • EAI硬體系統：「小腦」系統 - FPGA應用
    • EAI硬體系統：「小腦」系統 - MCU應用
    • EAI演算法：大腦控制技術路徑－視覺、語言和行為（VLA）模型
    • EAI演算法：腦控制技術路線－分層規劃架構
    • EAI演算法：腦控技術路線－跨機器人通用系統
    • EAI演算法：小腦控制技術路線－基於模型的控制方法
    • EAI演算法：小腦控制技術路線 - 模仿學習
    • EAI演算法：小腦控制的技術方法 - 深度強化學習
    • EAI演算法：大腦-小腦協調機制 - 傳統分層協調架構
    • EAI演算法：大腦-小腦協調機制－一種新型的類腦三系統結構（「大腦-腦橋-小腦」）
    • EAI硬體系統：機械系統
    • EAI硬體系統：機械系統（仿生骨架）
    • EAI機械系統：關節模組
    • EAI機械系統：聯合模組 - 馬達和積體電路
    • EAI機械系統：聯軸器模組 - 齒輪箱
    • EAI機械系統：關節模組 - 驅動器和編碼器
    • EAI硬體系統：執行系統
    • EAI硬體系統：執行系統（仿生肌肉）
    • EAI硬體系統：電源和溫度控管系統
    • EAI硬體系統：電源系統
    • EAI硬體系統：溫度控管系統
    • EAI硬體系統：感知系統
    • EAI硬體系統：感知系統
    • EAI硬體系統：感知系統框架
    • EAI硬體系統：感知系統 - 視覺感測器技術
    • EAI硬體系統：感知系統 - 雷達感測器技術
    • EAI硬體系統：感知系統 - 慣性測量單元（IMU）技術
  • EAI的軟體架構
  • EAI軟體架構簡介
  • EAI軟體架構：硬體抽象層（HAL）
  • EAI軟體架構：驅動程式執行層
  • EAI軟體架構：即時控制層
  • EAI軟體架構：決策層與規劃層
  • 企業應用整合（EAI）軟體架構：應用層（非即時層）
  • EAI 通訊架構
  • EAI通訊協定
  • EAI通訊協定：分層架構
  • EAI通訊協定：EtherCAT運行機制
  • EAI通訊協定：EtherCAT結構
  • EAI通訊協定：CAN運行機制
  • EAI通訊協定：CAN FD運行機制
  • EAI通訊協定：CAN FD網路框架
  • EAI評估標準
  • EAI 級別
  • 目前技術水準（1）
  • EAI目前的技術水準（2）
  • EAI目前的技術水準（3）

第3章：EAI 大腦（主控 SoC、控制器、大規模模型）

  • EAI主控SoC：機器人晶片和分組晶片概述
    • EAI主控SoC：機器人晶片組概覽 - 人形機器人
    • 人形機器人的主流設備端晶片和演算法
    • 人形機器人：Ubtech Walker S2、AgiBot Lingxi X2
    • 人形機器人：Unitree H2、Leju KUAVO 5
    • 人形機器人：Booster K1，Noetix Bumi
    • 人形機器人：EngineAI T800、ROBOTERA L7
    • 人形機器人：傅立葉智慧GR-3、小鵬IRON
    • 人形機器人：小米Cyber​​One、AI Figure 03
    • 人形機器人：特斯拉擎天柱第三代
    • 人形機器人：Noetix Hobbs 3（小諾）
    • EAI主控SoC：機器人晶片組概覽 - 四足機器人
    • 主流的四腳機器人晶片和演算法
    • 四足機器人：優利樹 AS2、小米 Cyber​​Dog
    • EAI 主控 SoC：機器人及分組晶片概述 - 其他機器人
    • 主流的機載晶片和演算法，適用於其他類型的機器人
    • 雙臂移動機器人：GigaAI Maker H01
  • EAI主控SoC：晶片供應商概述
  • EAI晶片供應商收入
  • EAI晶片供應商：半驅動產品列表
  • EAI晶片供應商：核心產品及半驅動器演進
  • EAI晶片供應商：NVIDIA產品列表
  • EAI晶片供應商：NVIDIA的關鍵產品和發展軌跡
  • EAI晶片供應商：高通產品列表
  • EAI晶片供應商：高通的核心產品與發展軌跡
  • EAI晶片供應商：英特爾產品列表
  • EAI晶片供應商：英特爾的關鍵產品與發展軌跡
  • EAI晶片供應商：聯發科產品列表
  • EAI晶片供應商：聯發科的關鍵產品與發展軌跡
  • EAI晶片供應商：瑞芯微產品列表
  • EAI晶片供應商：瑞芯微的關鍵產品與發展軌跡
  • EAI晶片供應商：黑芝麻科技產品列表
  • EAI晶片供應商：黑芝麻科技的核心產品及發展軌跡
  • EAI晶片供應商：寒武紀產品列表
  • EAI晶片供應商：寒武紀的關鍵產品和發展軌跡
  • EAI主控SoC的技術演進路徑
  • 趨勢 1
  • EAI 控制器：供應商摘要
  • EAI控制器：EAI控制器供應商的銷售額
  • EAI 控制器：SEER 機器人產品列表
  • EAI控制器：SEER Robotics的關鍵產品與發展歷程
  • EAI 控制器：iMotion
  • EAI控制器：立訊精密
  • EAI控制器：SIM技術
  • EAI控制器：成都瑞星星
  • EAI 控制器：NIIC
  • EAI控制器：飛馬座
  • EAI控制器：Inovance Technology
  • EAI控制器：華城工業控制
  • EAI大規模模式概述
    • EAI大型模型：VLA
    • 視覺-語言-行為（VLA）模型
    • VLA模型的起源：RT-1和RT-2
    • VLA模型的技術進步：OpenVLA
    • VLA模型的廣泛應用：圖示AI螺旋模型
    • VLA 型號應用範圍廣泛：NVIDIA GR00T N1
    • VLA模型應用範圍廣泛：位元組跳動GR-3模型
    • VLA模型的廣泛應用：Horizo​​n Robotics發布HoloBrain-0，一款全端開放原始碼VLA基礎模型
    • EAI大型模型：世界模型
    • 世界模型的基本架構
    • 世界模型的關鍵定義與應用發展
    • EAI世界模型概述
    • AgiBot 和上海人工智慧研究院提案了「EnerVerse」，這是一個他們共同創建的 4D 世界模型。
    • 3D-VLA：3D視覺、語言與行為生成世界模型
    • RoboDreamer：學習建構世界模型以增強機器人想像力
    • IRASim－世界機器人模型
    • Amap：ABot通用EAI系統
    • UnifoLM-WMA：Unitree開放原始碼世界模型
    • EAI模型的輕量級版本
    • 輕型機型部署的技術要求
    • 多模態融合與輕量化技術的結合
    • 輕量化技術：跨模態功能壓縮
    • 輕量化技術：動態模式選擇
    • 輕量化技術的應用：HugWBC通用人形機器人控制器
    • 輕量級技術的實現：HOVER多模態神經網路控制器
    • 輕量級技術實作：AMS（敏捷性和穩定性融合）框架

第4章 主流企業應用人工智慧機器人整合商

  • UBTECH
  • 產品和業務運營
  • 產品策略
  • 機器人SoC配置概述
  • 機器人模型演算法概述
  • 典型人形機器人參數比較
  • 人形機器人步行器S2：專用代理技術
  • 人形步行機器人S2：EAI大型模型思維機器人
  • 人形機器人步行機S2：自助式電池更換系統
  • 人形機器人步行器 S2：實現了端到端的類人立體視覺辨識。
  • 艾吉伯特
  • 輪廓
  • 機器人SoC配置概述
  • 模型演算法概述
  • 人形機器人參數比較
  • 人形機器人：實現平台模型 Genie Operator-1
  • 人形機器人：自主研發的控制系統
  • 人形機器人：AgiBot World－開放原始碼項目
  • 人形機器人：Powerflow核心關節模組和WITA互動式大型模型
  • 供應鏈
  • Unitree Robotics
  • 輪廓
  • 機器人SoC配置概述
  • 模型演算法概述
  • 四足機器人參數比較
  • 典型人形機器人參數比較
  • 消費者-向け四足機器人AS2：仿生大模型
  • As2，一款面向消費者的四足機器人：配備了專有的 4D LiDAR L2。
  • 供應鏈
  • 基本客群
  • Reju Robotics
  • 輪廓
  • 產品概述
  • 機器人SoC配置概述
  • 模型演算法概述
  • 機器人產品參數比較
  • 全端資料擷取與模型訓練系統
  • 樂居研究架構 2.0
  • 夥伴
  • 助推機器人
  • 輪廓
  • 機器人SoC配置概述
  • 機器人產品參數比較（1）
  • 機器人產品參數比較（2）
  • 諾埃蒂克機器人學
  • 輪廓
  • 機器人SoC配置概述
  • 模型演算法概述
  • 典型人形機器人參數比較
  • 仿生人形機器人參數比較
  • 專有的「Reikyu」運動控制演算法
  • 仿生機器人：專有的第二代仿生頭部平台
  • 專有的表達式驅動演算法和大規模多模態互動模型
  • EngineAI 機器人
  • 輪廓
  • 機器人SoC配置概述
  • 機器人產品參數比較
  • 運動控制演算法專利：Sim2Real Technology
  • 能源和結構專利
  • 聯合技術專利
  • 供應鏈
  • 機器人
  • 輪廓
  • 機器人SoC配置概述
  • 模型演算法概述
  • 機器人產品參數比較
  • Ctrl-世界世界模型
  • VLAW框架
  • 自主研發的端到端原生壓紋大型模型 ERA-42
  • ROBOTERA XHAND1 靈巧手
  • 供應鏈與成本結構：自主研發的核心零件 + 與策略供應商的合作
  • 傅立葉智慧
  • 輪廓
  • 機器人SoC配置概述
  • 典型人形機器人參數比較
  • FSA 2.0 執行器
  • 伽利略系統
  • 超人工智慧
  • 輪廓
  • 產品參數
  • 超級大腦
  • 超世界
  • Xpeng IRON
  • 輪廓
  • IRON機器人：商業化進展與未來計劃
  • IRON人形機器人：產品參數對比
  • IRON機器人主控SoC：專有圖靈AI晶片
  • IRON機器人主控SoC：自主研發的圖靈AI晶片的詳細參數
  • IRON機器人主控SoC：專有圖靈AI晶片的參數解讀
  • IRON 機器人 AI 大規模模型：第二代 VLA 物理世界大規模模型的應用
  • IRON 機器人雲端基礎模式：可與汽車系統重複使用
  • IRON機器人感知系統：鷹眼視覺系統
  • IRON機器人成本與供應鏈結構：第一代IRON的成本
  • 小米
  • Cyber​​One機器人參數
  • Cyber​​Dog四足機器人參數
  • 機器人：VLA 基礎模型 Xiaomi-Robotics-0
  • 機器人：專有軟體演算法
  • 機器人：Cyber​​One仿生手
  • 機器人：專有動力系統
  • 機器人技術：成本與供應鏈配置
  • 機器人技術：商業化進展及未來規劃
  • 特斯拉
  • Tesla Optimus 參數
  • 面向人形機器人的主流設備端運算晶片：特斯拉 A15
  • 特斯拉Optimus Gen 3的運動控制：基於Dojo超級電腦訓練的強化學習模型
  • 特斯拉 Optimus Gen 3：重複使用 FSD V12/V13 視覺專用神經網路架構
  • Tesla Optimus Gen 3：運動規劃演算法
  • 特斯拉Optimus Gen 3：靈巧的手
  • 特斯拉Optimus供應鏈
  • 圖 AI
  • 輪廓
  • 機器人SoC配置和模型演算法概述
  • 典型人形機器人參數比較
  • 機器人：Helix AI模型
  • 機器人：BotQ人形機器人工廠
  • 供應鏈

第5章：主要企業應用整合晶片供應商

  • 半驅動
  • EAI產品的應用與規劃
  • 戰略 2.0：從駕駛情報到通用情報
  • EAI「Cerebrum」SoC的詳細參數
  • EAI「小腦」SoC的詳細參數
  • EAI「大腦」SoC：R1
  • 智慧控制小腦SoC：D9-MAX
  • 智慧控制小腦SoC D9-MAX：應用解決方案與開發套件
  • 用於EAI的高性能MCU的詳細參數
  • 基於E3119的聯合模組解決方案
  • 基於E3116的便利解決方案
  • 基於E3118的LiDAR解決方案
  • 鎖晶片
  • 輪廓
  • EAI晶片的演進與未來發展
  • RK3588系列產品規格
  • RK182X協處理器SoC和RV1126B影像處理器的參數
  • RK182X系列協處理器SoC及應用解決方案
  • RK3588
  • RK3588系列產品的應用解決方案與未來規劃
  • RK3588 應用解決方案：研華增強型視覺控制器
  • RK3588 應用解決方案：高效能 AMR 機器人核心運算平台解決方案
  • RK3588 開發工具鏈：RKNN-Toolkit2
  • D-Robotics
  • EAI晶片的演進與未來發展
  • EAI SoC產品參數
  • EAI 開發套件產品參數
  • 日昇5智慧運算晶片，CPU+BPU異構架構
  • 智慧型運算晶片應用生態系統：NIU電動摩托車智慧運輸
  • 開發者套件應用生態系統：SENSINGTech 的 GMSL2 系列相機模組
  • 黑芝麻科技
  • EAI晶片的演進與未來發展
  • 華山A2000
  • SesameX EAI 運算平台模組
  • 華山A2000
  • SesameX：全端機器人平台系統
  • 寒武紀
  • EAI晶片的演進與未來發展
  • EAI晶片的詳細參數
  • 思源590：專有智慧處理器微架構MLUarch05
  • 人工智慧運算庫：Cambricon CNNL
  • 電腦視覺庫：CNCV
  • 軟體開發平台：Campricon NeuWare
  • MLU推理加速引擎：MagicMind
  • NVIDIA
  • 輪廓
  • EAI SoC系列及其演進
  • 面向人形機器人的主流設備端運算晶片：Jetson Orin
  • Jetson Orin 詳細參數
  • 面向人形機器人的主流設備端運算晶片：Jetson Thor
  • Jetson Thor 的詳細參數
  • NVIDIA Jetson Thor：其GPU採用Blackwell架構。
  • NVIDIA Jetson Thor：視覺 AI 代理的 NVIDIA Metropolis
  • NVIDIA Jetson Thor：支援 NVIDIA Holoscan 感測器處理的即時資料流傳輸。
  • NVIDIA Jetson Thor：JetPack 7 提供了一套完整的工具和函式庫，用於建立 AI 邊緣應用程式。
  • NVIDIA Jetson Thor：與開放原始碼機器人平台「Isaac」的整合
  • NVIDIA DreamZero 世界動作模型 (WAM)
  • NVIDIA DreamZero 世界動作模型 (WAM)：架構
  • NVIDIA DreamZero 世界動作模型 (WAM)：優勢
  • 開放式多模態模型：Nemotron 3 Nano Omni 模型
  • 高通
  • EAI晶片的演進與未來發展
  • Dragonwing系列晶片的詳細參數：IQ10、IQ9
  • Dragonwing系列晶片（IQ8、IQ6、QCS8550）的詳細規格
  • IQ10系列
  • QCS8550 應用解決方案：Robrain AI 機器人解決方案
  • 英特爾
  • EAI SoC系列及其演進
  • Core Ultra系列產品參數對比
  • 英特爾酷睿i7系列詳細規格
  • 英特爾酷睿i5系列詳細規格
  • 用於內建設備的機器人的計算晶片：第三代英特爾酷睿Ultra處理器
  • 第三代英特爾® 酷睿™ Ultra：18A 處理器
  • 第三代英特爾酷睿Ultra GPU架構：Xe3
  • 第三代英特爾® 酷睿™ 超級處理器（附 NPU 5）：專為 AI 任務最佳化
  • 聯發科
  • EAI晶片的演進與未來發展
  • Genio Pro、Genio 420、Genio 360
  • 天璣9400，天璣9400+
  • Genio Pro
  • Genio 420
  • Genio 360
  • 支援聯發科 NeuroPilot AI 軟體開發工具包
  • Li Auto
  • 馬赫數 M100 參數
  • 自主研發的晶片，Mach M100
  • 專有晶片「Mach M100」：內部結構
  • Mach M100專用晶片：CPU結構
  • Mach M100專有晶片：NPU結構
  • HOUMO.AI
  • 具身智慧晶片的演進與未來前景
  • 厚膜曼傑M50晶片（1片）
  • 厚膜曼傑M50晶片（2片）
  • Houmo Manjie M50：配備專有的第二代記憶體運算IPU設計，「天軒」架構。
  • 厚膜曼傑M50工具鏈：厚膜大道
  • 匯熙智慧科技
  • 具身智慧晶片的演進與未來前景
  • Hoi C R1
  • 專有的圖靈完備指令集
  • 專有的 RPU 神經網路加速器
  • 創新功能安全架構 RIF
簡介目錄
Product Code: XX019

Embodied AI SoC Research: Chip Vendors Are Transforming from "Single SoC Vendors" to "Full-Stack Chip Platform Providers".

The advancing chip technology provides a crucial boost to the booming embodied artificial intelligence (EAI) industry. Robots for different application scenarios have differentiated chip selection requirements, avoiding problems caused by improper selection, such as excessive computing power surplus or insufficient performance. In addition, the development of the EAI industry also relies on breakthroughs in large model technology. The intelligence level of robots has been significantly improved, enabling robots to make independent judgments and perform complex tasks.

As the EAI Market Continues to Expand and Chip Performance Requirements Keeps Rising, Chip Vendors Launch Full-Stack Solutions.

The robot chip market is in a period of rapid growth. The global shipments of general-purpose EAI robots reached 13,000 units in 2025 and is expected to exceed 50,000 units in 2026. At present, major chip giants have launched SoCs for EAI, such as NVIDIA Jetson series and Qualcomm IQ10 series. Meanwhile, they provide robot development platforms, including NVIDIA Isaac open-source platform, second-generation Rockchip RKNN neural network model conversion and optimization tool, and Black Sesame SmartX multi-dimensional intelligent computing platform for the robot industry, in a bid to meet customers' needs for rapid application deployment and model development.

Currently, EAI SoCs are evolving:

Trend 1: Requirements for Chip Computing Power Become Much Higher.

NVIDIA's new Jetson T5000 adopts the Blackwell GPU architecture, delivering up to 2070 FP4 TFLOPS of AI compute, 7.5x higher than the previous-generation Jetson Orin. The RDK S100P from Horizon Robotics (D-Robotics) integrates CPU+BPU+MCU on a single chip, delivering 120 TOPS computing power. With the increasing complexity of algorithms, robots' computing power demand is gradually rising from the current 200-500 TOPS to 500-1000 TOPS. Notably, the industry no longer simply stacks computing power but shifts to "efficiency priority". Algorithm optimization makes efficiency a core indicator.

Trend 2: Chip Vendors Evolve towards Advanced Processes

Mainstream chip vendors move towards advanced processes. NVIDIA Jetson AGX Thor adopts 4nm process, Intel Core Ultra Series 3 uses Intel 18A process, Rockchip RK3588 adopts 8nm process, and MediaTek's latest Genio Pro adopts 3nm process, substantially boosting chip performance.

Trend 3: EAI Chip Vendors Are Transforming from "Single SoC Vendors" To "Full-Stack Chip Platform Providers".

In SemiDrive's case, besides EAI cerebrum SoCs, it has also launched intelligent control cerebellum SoCs and high-performance MCUs, so as to build full-stack EAI solutions, covering the complete architecture of "cerebrum - cerebellum - body- joint". Its product matrix ranges from main control cerebrum SoCs for high-level cognition and decision, and intelligent control cerebellum chips for motion coordination and real-time control, to E3-R series MCUs for LiDAR/machine vision, motion center, dexterous hands and joint modules, realizing full-chain chip coverage.

Among them, the intelligent control cerebellum D9-Max and robot joint module MCU E311x-R have come into mass production, and built in-depth cooperation with leading robot enterprises, successfully bringing automotive-grade high performance and high reliability into the robot field.

D9 Max adopts an architecture optimized for cerebellum application. Based on hardware isolation and hardware virtualization technology, it integrates one 8-core 2.0GHz Cortex-A55 CPU cluster, one 4-core 2.0GHz Cortex-A55 CPU cluster, and 3 pairs of dual-core lockstep 800MHz Cortex-R5F, as well as computing units like 8TOPS NPU and GPU. A single chip allows for deployment of three core functions of motion control system, HMI and EtherCAT master station, integrating the functions that traditional solutions require three chips to enable into a single chip.

The high-performance MCUs (E3-R series) have made substantial progress in joint control, meeting high functional safety and cybersecurity requirements and providing one-stop solutions. As the main control chip for joint modules, E311x-R features high real-time performance and highly stable computing power output capability. It adopts dual R5F cores with a main frequency up to 400MHz. In actual R&D, the dual cores separate motor control and communication processing for dedicated core allocation and enhanced performance.

In terms of EAI cerebrum SoC, SemiDrive reuses its expertise in on-device large model capabilities in the automotive sector to develop the next-generation robot cerebrum chip R1. Adopting ARM V9.2 architecture CPU and new high-performance NPU, it supports on-device deployment of embodied end-to-end models such as MLLM/VLA under low power consumption.

Trend 4: Chip Vendors Are Launching Full-stack Self-developed Toolchains.

Rockchip launched RKNN-Toolkit2, its second-generation neural network model conversion and optimization tool. Acting as a bridge connecting mainstream deep learning frameworks and Rockchip NPU (Neural Processing Unit) hardware platforms, it is designed to help developers efficiently deploy trained AI models on embedded devices. Based on Huashan A2000, Black Sesame Technologies builds the easy-to-use Shanhai AI toolchain, covering the entire process from model optimization to on-device deployment, providing developers with an efficient model development and deployment system. SemiDrive offers complete software and hardware development kits such as the D9-Max application development kit, enabling customers and independent developers to rapidly deploy applications and conduct on-device development.

Selection of Chips and Algorithms by Embodied Robot OEMs

The EAI level is essentially the result of the co-evolution of algorithms and chips. The two are interdependent and mutually driven, forming the core closed loop of robot intelligent systems.

For example, the basic computing board of AgiBot Lingxi X2 adopts two Rockchip RK3588 chips, replacing Jetson Xavier adopted by the previous-generation, offering improvements in both cost and performance. The 6TOPS NPU of RK3588 delivers excellent performance in motion control and perception fusion scenarios while reducing power consumption by 7W. The high-compute board adopts NVIDIA Jetson Orin NX, with total AI compute of 169 TOPS.

In terms of algorithms, the cerebrum of Lingxi X2 is equipped with AgiBot's self-developed large model Genie Operator-1 (GO-1). Adopting the Vision-Language-Latent-Action (ViLLA) architecture composed of VLM (multimodal large model) and mixture-of-experts (MoE), Lingxi X2 possesses superior learning, fast few-shot generalization and continuous evolution capabilities. The cerebellum of Lingxi X2 adopts the Xyber-Edge controller for robot motion coordination and decision. With a 144-core heterogeneous computing architecture, the controller dynamically allocates reasoning tasks to NPU clusters, and control commands to FPGAs, and compresses the traditional 12-layer control architecture for motion planning into a 3-layer implicit planning structure, achieving 450Hz real-time closed-loop control, greatly superior to Tesla Optimus' 280Hz closed-loop frequency.

AgiBot has made a differentiated and complementary layout by launching three product series of Yuanzheng, Lingxi and Genie, targeting industrial manufacturing, commercial services and data research scenarios, respectively, and is advancing towards mass production and commercial deployment.

Table of Contents

1 EAI Market and Application Scenarios

  • 1.1 Basic Concepts and Terminology of EAI
  • Basic Concepts of EAI (1)
  • Basic Concepts of EAI (2)
  • Basic Concepts of EAI (3)
  • Terminology of EAI (1)
  • Terminology of EAI (2)
  • 1.2 Market Prospect of EAI
  • Evolution History of EAI
  • Status Quo of EAI Industry
  • Evolution of EAI Application Scenarios (1)
  • Evolution of EAI Application Scenarios (2)
  • EAI Market Trends
  • China EAI Market Size
  • Global Humanoid Robot Shipments
  • 1.3 Application Prospect of EAI
  • Summary of Application Prospects
  • EAI Market Structure by Application Scenario
  • Community & Household Scenario: Household Service (1)
  • Community & Household Scenario: Household Service (2)
  • Community & Household Scenario: Medical / Nursing Scenario (1)
  • Community & Household Scenario: Medical / Nursing Scenario (2)
  • Smart Manufacturing Scenario: Factory Production
  • Smart Manufacturing Scenario: Figure Humanoid Robots Realize 24/7 Operation in Factories
  • Smart Manufacturing Scenario: UBTECH Walker S2 Group Collaborative Operation in Smart Factories
  • Smart Manufacturing Scenario: Agricultural Production
  • Commercial Service Scenario: KEENON Robotics
  • Commercial Service Scenario: Meituan "Little Wasp"
  • High-Risk & Rescue Scenario: DEEP Robotics LYNX M20 Wheeled-Legged Robot
  • High-Risk & Rescue Scenario: iFreecomm "Lingmu" Emergency Rescue Quadruped Robot and Guide Dog
  • 1.4 Competition Summary of EAI Suppliers
  • Top 50 Chinese EAI Suppliers (1)
  • Top 50 Chinese EAI Suppliers (2)
  • Top 10 Foreign EAI Suppliers (1)
  • Global Shipments of Top 10 Humanoid Robots, 2025 (Mainstream Statistical Caliber)
  • Revenues of Representative EAI Enterprises (1)
  • Revenues of Representative EAI Enterprises (2)
  • Technical Routes of Representative EAI Enterprises (1)
  • Technical Routes of Representative EAI Enterprises (2)
  • Technical Routes of Representative EAI Enterprises (3)

2 Software and Hardware System Architecture of EAI

  • 2.1 Hardware Architecture of EAI
  • EAI: Introduction to Hardware System
  • EAI Hardware List
  • EAI Chip List
  • SemiDrive: Full-Stack Chip Solutions for Robots
  • GigaDevice: Full-Stack Chip Solutions for Robots
  • Infineon: Solutions for Each Functional Module of Humanoid Robots (1)
  • Infineon: Solutions for Each Functional Module of Humanoid Robots (2)
  • Infineon: Product Layout for Humanoid Robots
    • 2.1.1 EAI Hardware System: Computing Power and Hardware Control System
    • EAI Hardware System: Computing Power and Hardware Control System
    • EAI Hardware System: Composition of the "Cerebrum" System
    • EAI Hardware System: "Cerebrum" System - Application of Main Control SoC
    • EAI Hardware System: "Cerebellum" System - Application of FPGA
    • EAI Hardware System: "Cerebellum" System - Application of MCU
    • EAI Algorithm: Cerebrum Control Technical Route - Vision-Language-Action (VLA) Model
    • EAI Algorithm: Cerebrum Control Technical Route - Hierarchical Planning Architecture
    • EAI Algorithm: Cerebrum Control Technical Route - Cross-Robot General System
    • EAI Algorithm: Cerebellum Control Technical Route - Model-Based Control Method
    • EAI Algorithm: Cerebellum Control Technical Route - Imitation Learning
    • EAI Algorithm: Cerebellum Control Technical Route - Deep Reinforcement Learning
    • EAI Algorithm: Cerebrum-Cerebellum Collaboration Mechanism - Traditional Hierarchical Collaboration Architecture
    • EAI Algorithm: Cerebrum-Cerebellum Collaboration Mechanism - New Brain-Inspired Three-System Architecture ("Cerebrum - Pons - Cerebellum")
    • 2.1.2 EAI Hardware System: Mechanical System
    • EAI Hardware System: Mechanical System (Bionic Skeleton)
    • EAI Mechanical System: Joint Module
    • EAI Mechanical System: Joint Module - Motor and IC
    • EAI Mechanical System: Joint Module - Reducer
    • EAI Mechanical System: Joint Module - Driver and Encoder
    • 2.1.3 EAI Hardware System: Execution System
    • EAI Hardware System: Execution System (Bionic Muscle)
    • 2.1.4 EAI Hardware System: Power Supply and Thermal Management System
    • EAI Hardware System: Power Supply System
    • EAI Hardware System: Thermal Management System
    • 2.1.5 EAI Hardware System: Perception System
    • EAI Hardware System: Perception System
    • EAI Hardware System: Perception System Framework
    • EAI Hardware System: Perception System - Vision Sensor Technology
    • EAI Hardware System: Perception System - Radar Sensor Technology
    • EAI Hardware System: Perception System - Inertial Measurement Unit (IMU) Technology
  • 2.2 Software Architecture of EAI
  • Introduction to EAI Software Architecture
  • EAI Software Architecture: Hardware Abstraction Layer (HAL)
  • EAI Software Architecture: Driver Execution Layer
  • EAI Software Architecture: Real-Time Control Layer
  • EAI Software Architecture: Decision & Planning Layer
  • EAI Software Architecture: Application Layer (Non-Real-Time Layer)
  • 2.3 Communication Architecture of EAI
  • Communication Protocol of EAI
  • Communication Protocol of EAI: Hierarchical Architecture
  • Communication Protocol of EAI: Working Mechanism of EtherCAT
  • Communication Protocol of EAI: Structure of EtherCAT
  • Communication Protocol of EAI: Working Mechanism of CAN
  • Communication Protocol of EAI: Working Mechanism of CAN FD
  • Communication Protocol of EAI: CAN FD Network Framework
  • 2.4 Grading Standard for EAI
  • Levels of EAI
  • Current Technical Level of EAI (1)
  • Current Technical Level of EAI (2)
  • Current Technical Level of EAI (3)

3 EAI Cerebrum (Main Control SoC, Controller and Large Model)

  • 3.1 EAI Main Control SoC: Summary of Robots and Grouped Chips
    • 3.1.1 EAI Main Control SoC: Summary of Robots and Grouped Chips -Humanoid Robots
    • Mainstream On-device Chips and Algorithms for Humanoid Robots
    • Humanoid Robots: Ubtech Walker S2, AgiBot Lingxi X2
    • Humanoid Robots: Unitree H2, Leju KUAVO 5
    • Humanoid Robots: Booster K1, Noetix Bumi
    • Humanoid Robots: EngineAI T800, ROBOTERA L7
    • Humanoid Robots: Fourier Intelligence GR-3, Xpeng IRON
    • Humanoid Robots: Xiaomi CyberOne, Figure AI Figure 03
    • Humanoid Robots: Tesla Optimus Gen 3
    • Humanoid Robots: Noetix Hobbs 3 (Xiaonuo)
    • 3.1.2 EAI Main Control SoC: Summary of Robots and Grouped Chips -Quadruped Robots
    • Mainstream On-device Chips and Algorithms for Quadruped Robots
    • Quadruped Robots: Unitree As2, Xiaomi CyberDog
    • 3.1.3 EAI Main Control SoC: Summary of Robots and Grouped Chips - Other Robots
    • Mainstream On-device Chips and Algorithms for Other Types of Robots
    • Dual-Arm Mobile Robot: GigaAI Maker H01
  • 3.2 EAI Main Control SoC: Summary of Chip Vendors
  • Revenues of EAI Chip Vendors
  • EAI Chip Vendors: Product List of SemiDrive
  • EAI Chip Vendors: Core Products and Evolution Route of SemiDrive
  • EAI Chip Vendors: Product List of NVIDIA
  • EAI Chip Vendors: Core Products and Evolution Route of NVIDIA
  • EAI Chip Vendors: Product List of Qualcomm
  • EAI Chip Vendors: Core Products and Evolution Route of Qualcomm
  • EAI Chip Vendors: Product List of Intel
  • EAI Chip Vendors: Core Products and Evolution Route of Intel
  • EAI Chip Vendors: Product List of MediaTek
  • EAI Chip Vendors: Core Products and Evolution Route of MediaTek
  • EAI Chip Vendors: Product List of Rockchip
  • EAI Chip Vendors: Core Products and Evolution Route of Rockchip
  • EAI Chip Vendors: Product List of Black Sesame Technologies
  • EAI Chip Vendors: Core Products and Evolution Route of Black Sesame Technologies
  • EAI Chip Vendors: Product List of Cambricon
  • EAI Chip Vendors: Core Products and Evolution Route of Cambricon
  • 3.3 Technical Evolution Route of EAI Main Control SoC
  • Trend 1:
  • Trend 2:
  • Trend 3:
  • 3.4 EAI Controller: Summary of Suppliers
  • EAI Controller: Revenues of EAI Controller Suppliers
  • EAI Controller: Product List of SEER Robotics
  • EAI Controller: Core Products and Evolution Route of SEER Robotics
  • EAI Controller: IMotion
  • EAI Controller: Luxshare Precision
  • EAI Controller: SIM Technology
  • EAI Controller: Chengdu Ruixingxing
  • EAI Controller: NIIC
  • EAI Controller: Pegasus?
  • EAI Controller: Inovance Technology
  • EAI Controller: Huacheng Industrial Control
  • 3.5 Summary of EAI Large Models
    • 3.5.1 EAI Large Model: VLA
    • Vision-Language-Action (VLA) Model
    • Origin of VLA Model: RT-1 and RT-2
    • Technical Deepening of VLA Model: OpenVLA
    • Wide Application of VLA Model: Figure AI Helix Model
    • Wide Application of VLA Model: NVIDIA GR00T N1
    • Wide Application of VLA Model: ByteDance GR-3 Model
    • Wide Application of VLA Model: Horizon Robotics Released Full-Stack Open-Source VLA Foundation Model HoloBrain-0
    • 3.5.2 EAI Large Model: World Model
    • Basic Architecture of World Model
    • Key Definition and Application Development of World Model
    • Summary of EAI World Models
    • AgiBot and Shanghai AI Lab Jointly Proposed Embodied 4D World Model EnerVerse
    • 3D-VLA: A 3D Vision-Language-Action Generative World Model
    • RoboDreamer: Learning Compositional World Models for Robot Imagination
    • IRASim - World Model in Robotics
    • Amap: ABot General EAI System (1)
    • Amap: ABot General EAI System (2)
    • UnifoLM-WMA: Unitree Open-Source World Model
    • 3.5.3 Lightweight Deployment of EAI Models
    • Technical Requirements for Lightweight Model Deployment
    • Combination of Multimodal Fusion and Lightweight Technology
    • Lightweight Technology: Cross-Modal Feature Compression
    • Lightweight Technology: Dynamic Modal Selection
    • Lightweight Technology Implementation: HugWBC General Humanoid Robot Controller
    • Lightweight Technology Implementation: HOVER Multimodal Neural Network Controller
    • Lightweight Technology Implementation: AMS (Agility Meets Stability) Framework

4 Mainstream EAI Robot Integrators

  • 4.1 UBTECH
  • Products and Operation
  • Product Strategy
  • Overview of Robot SoC Configurations
  • Overview of Robot Model Algorithms
  • Parameter Comparison between General Humanoid Robots (1)
  • Parameter Comparison between General Humanoid Robots (2)
  • Parameter Comparison between General Humanoid Robots (3)
  • Humanoid Robot Walker S2: Dedicated Agent Technology
  • Humanoid Robot Walker S2: EAI Large Model Thinker
  • Humanoid Robot Walker S2: Self-Service Battery Swap System
  • Humanoid Robot Walker S2: End-to-End Human-Like Stereo Vision Perception
  • 4.2 AgiBot
  • Profile
  • Overview of Robot SoC Configurations (1)
  • Overview of Robot SoC Configurations (2)
  • Overview of Model Algorithms
  • Parameter Comparison between Humanoid Robots (1)
  • Parameter Comparison between Humanoid Robots (2)
  • Parameter Comparison between Humanoid Robots (3)
  • Humanoid Robot: Embodied Foundation Model Genie Operator-1
  • Humanoid Robot: Self-Developed Controller System
  • Humanoid Robot: Million-Level Real Robot Dataset Open-Source Project AgiBot World
  • Humanoid Robot: Powerflow Core Joint Module and WITA Interactive Large Model
  • Supply Chain (1)
  • Supply Chain (2)
  • 4.3 Unitree Robotics
  • Profile
  • Overview of Robot SoC Configurations (1)
  • Overview of Robot SoC Configurations (2)
  • Overview of Model Algorithms
  • Parameter Comparison between Quadruped Robots (1)
  • Parameter Comparison between Quadruped Robots (2)
  • Parameter Comparison between Quadruped Robots (3)
  • Parameter Comparison between Quadruped Robots (4)
  • Parameter Comparison between General Humanoid Robots (1)
  • Parameter Comparison between General Humanoid Robots (2)
  • Parameter Comparison between General Humanoid Robots (3)
  • Consumer-grade Quadruped Robot As2: Bionic Embodied Large Model
  • Consumer-grade Quadruped Robot As2: Self-Developed 4D LiDAR L2
  • Supply Chain
  • Customer Base
  • 4.4 Leju Robotics
  • Profile
  • Product Overview
  • Overview of Robot SoC Configurations
  • Overview of Model Algorithms
  • Parameter Comparison between Robot Products (1)
  • Parameter Comparison between Robot Products (2)
  • Parameter Comparison between Robot Products (3)
  • Full-Stack Data Collection and Model Training System
  • Leju Research Framework 2.0 (1)
  • Leju Research Framework 2.0 (2)
  • Partners
  • 4.5 Booster Robotics
  • Profile
  • Overview of Robot SoC Configurations
  • Parameter Comparison between Robot Products (1)
  • Parameter Comparison between Robot Products (2)
  • 4.6 Noetix Robotics
  • Profile
  • Overview of Robot SoC Configurations
  • Overview of Model Algorithms
  • Parameter Comparison between General Humanoid Robots (1)
  • Parameter Comparison between General Humanoid Robots (2)
  • Parameter Comparison between Bionic Humanoid Robots (1)
  • Parameter Comparison between Bionic Robot Products (2)
  • Self-Developed "Lingjiu" Motion Control Algorithm
  • Bionic Robot: Self-Developed Second-Generation Bionic Head Platform
  • Self-Developed Expression Driven Algorithm and Multimodal Interaction Large Model
  • 4.7 EngineAI Robotics
  • Profile
  • Overview of Robot SoC Configurations of
  • Parameter Comparison between Robot Products (1)
  • Parameter Comparison between Robot Products (2)
  • Motion Control Algorithm Patent: Sim2Real Technology
  • Energy and Structural Patents
  • Joint Technology Patents
  • Supply Chain
  • 4.8 ROBOTERA
  • Profile
  • Overview of Robot SoC Configurations
  • Overview of Model Algorithms
  • Parameter Comparison between Robot Products (1)
  • Parameter Comparison between Robot Products (2)
  • Ctrl-World World Model
  • VLAW Framework
  • Self-Developed Native End-to-End Embodied Large Model ERA-42
  • ROBOTERA XHAND1 Dexterous Hand
  • Supply Chain and Cost Composition: Self-Developed Core Components + Cooperation with Strategic Suppliers
  • 4.9 Fourier Intelligence
  • Profile
  • Overview of Robot SoC Configurations
  • Parameter Comparison between General Humanoid Robots (1)
  • Parameter Comparison between General Humanoid Robots (2)
  • Parameter Comparison between General Humanoid Robots (3)
  • FSA 2.0 Actuator
  • Galileo System
  • 4.10 GigaAI
  • Profile
  • Product Parameters
  • GigaBrain
  • GigaWorld
  • 4.11 Xpeng IRON
  • Profile
  • IRON Robot: Commercialization Progress and Future Planning
  • IRON Humanoid Robot: Product Parameter Comparison (1)
  • IRON Humanoid Robot: Product Parameter Comparison (2)
  • IRON Humanoid Robot: Product Parameter Comparison (3)
  • IRON Humanoid Robot: Product Parameter Comparison (4)
  • IRON Humanoid Robot: Product Parameter Comparison (5)
  • IRON Robot Main Control SoC: Self-Developed Turing AI Chip
  • IRON Robot Main Control SoC: Detailed Parameters of Self-Developed Turing AI Chip
  • IRON Robot Main Control SoC: Parameter Interpretation of Self-Developed Turing AI Chip
  • IRON Robot AI Large Model: Application of Second-Generation VLA Physical World Large Model
  • IRON Robot Cloud Foundation Model: Reusable with Automobiles
  • IRON Robot Perception System: Hawk-Eye Vision System
  • IRON Robot Cost and Supply Chain Composition: Cost of the First-Generation IRON
  • 4.12 Xiaomi
  • Parameters of CyberOne Robot (1)
  • Parameters of CyberOne Robot (2)
  • Parameters of CyberOne Robot (3)
  • Parameters of CyberDog Quadruped Robot
  • Robot: VLA Foundation Model Xiaomi-Robotics-0 (1)
  • Robot: VLA Foundation Model Xiaomi-Robotics-0 (2)
  • Robot: Self-Developed Software Algorithm
  • Robot: CyberOne Bionic Hand (1)
  • Robot: CyberOne Bionic Hand (2)
  • Robot: Self-Developed Power System
  • Robot: Cost and Supply Chain Composition
  • Robot: Commercialization Progress and Future Planning
  • 4.13 Tesla
  • Parameters of Tesla Optimus (1)
  • Parameters of Tesla Optimus (2)
  • Parameters of Tesla Optimus (3)
  • Mainstream On-device Computing Chip for Humanoid Robots: Tesla A15
  • Tesla Optimus Gen 3 Motion Control: Reinforcement Learning Model Trained by Dojo Supercomputer
  • Tesla Optimus Gen 3: Reuse FSD V12/V13 Vision-only Neural Network Architecture (1)
  • Tesla Optimus Gen 3: Reuse FSD V12/V13 Vision-only Neural Network Architecture (2)
  • Tesla Optimus Gen 3: Reuse FSD V12/V13 Vision-only Neural Network Architecture (3)
  • Tesla Optimus Gen 3: Reuse FSD V12/V13 Vision-only Neural Network Architecture (4)
  • Tesla Optimus Gen 3: Reuse FSD V12/V13 Vision-only Neural Network Architecture (5)
  • Tesla Optimus Gen 3: Motion Planning Algorithm
  • Tesla Optimus Gen 3: Dexterous Hand (1)
  • Tesla Optimus Gen 3: Dexterous Hand (2)
  • Tesla Optimus Gen 3: Dexterous Hand (3)
  • Supply Chain of Tesla Optimus
  • 4.14 Figure AI
  • Profile
  • Overview of Robot SoC Configurations and Model Algorithms
  • Parameter Comparison between General Humanoid Robots
  • Robot: Helix AI Model
  • Robot: BotQ Humanoid Robot Factory
  • Supply Chain

5 Mainstream EAI Chip Vendors

  • 5.1 SemiDrive
  • Application and Planning of EAI Products
  • Strategy 2.0 from Driving Intelligence to General Intelligence
  • Detailed Parameters of EAI "Cerebrum" SoC
  • Detailed Parameters of EAI "Cerebellum" SoC
  • EAI "Cerebrum" SoC: R1
  • Intelligent Control Cerebellum SoC: D9-MAX
  • Intelligent Control Cerebellum SoC D9-MAX: Application Solution and Development Kit
  • Detailed Parameters of High-Performance MCU for EAI
  • Joint Module Solution Based on E3119
  • Dexterous Hand Solution Based on E3116
  • LiDAR Solution Based on E3118
  • 5.2 Rockchip
  • Profile
  • Evolution and Future Development of EAI Chips
  • Parameters of RK3588 Series Products
  • Parameters of RK182X Co-processor SoC & RV1126B Image Processor
  • RK182X Series Co-Processor SoCs and Application Solutions
  • RK3588
  • RK3588 Series Application Solution and Future Planning
  • RK3588 Application Solution: Advantech?Reinforced Vision Controller
  • RK3588 Application Solution: High-Performance AMR Robot Core Computing Platform Solution
  • RK3588 Development Toolchain: RKNN-Toolkit2
  • 5.3 D-Robotics
  • Evolution and Future Development of EAI Chips
  • Parameters of EAI SoC Products
  • Parameters of EAI Developer Kit Product
  • Sunrise 5 Intelligent Computing Chip, CPU+BPU Heterogeneous Architecture
  • Intelligent Computing Chip Application Ecosystem: NIU Electric Two-Wheeler Smart Mobility
  • Developer Kit Application Ecosystem: SENSING?Tech's GMSL2 Series Camera Module
  • 5.4 Black Sesame Technologies
  • Evolution and Future Development of EAI Chips
  • Huashan A2000 (1)
  • Huashan A2000 (2)
  • SesameX EAI Computing Platform Module
  • Huashan A2000
  • Huashan A2000: Adopt Self-Developed Jiushao Architecture NPU Core
  • Huashan A2000: Efficient, Easy-to-use Shanhai AI Toolchain
  • SesameX: Full-Stack Robot Platform System
  • 5.5 Cambricon
  • Evolution and Future Development of EAI Chips
  • Detailed Parameters of EAI Chips (1)
  • Detailed Parameters of EAI Chips (2)
  • Siyuan 590: Self-Developed Intelligent Processor Microarchitecture MLUarch05
  • AI Computing Library: Cambricon CNNL
  • Computer Vision Library: CNCV
  • Software Development Platform: Cambricon NeuWare
  • MLU Inference Acceleration Engine: MagicMind
  • 5.6 NVIDIA
  • Profile
  • EAI SoC Series and Evolution
  • Mainstream On-device Computing Chip for Humanoid Robots: Jetson Orin
  • Detailed Parameters of Jetson Orin
  • Mainstream On-device Computing Chip for Humanoid Robots: Jetson Thor
  • Detailed Parameters of Jetson Thor
  • NVIDIA Jetson Thor: Adopt Blackwell Architecture for GPU
  • NVIDIA Jetson Thor: NVIDIA Metropolis for Vision AI Agents
  • NVIDIA Jetson Thor: NVIDIA Holoscan for Sensor Processing to Realize Real-Time Data Stream Transmission
  • NVIDIA Jetson Thor: JetPack 7 Provides Complete Tools and Libraries for Building AI Edge Applications
  • NVIDIA Jetson Thor: Collaborate with Isaac Open-Source Robot Platform
  • NVIDIA DreamZero World Action Model (WAM)
  • NVIDIA DreamZero World Action Model (WAM): Architecture
  • NVIDIA DreamZero World Action Model (WAM): Advantages
  • Open Multimodal Model: Nemotron 3 Nano Omni Model
  • 5.7 Qualcomm
  • Evolution and Future Development of EAI Chips
  • Detailed Parameters of Dragonwing Series Chips: IQ10, IQ9
  • Detailed Parameters of Dragonwing Series Chips: IQ8, IQ6, QCS8550
  • IQ10 Series
  • QCS8550 Application Solution: Robrain AI Robot Solution
  • 5.8 Intel
  • EAI SoC Series and Evolution
  • Parameter Comparison between Core Ultra Series Products
  • Detailed Parameters of Intel Core i7 Series
  • Detailed Parameters of Intel Core i5 Series
  • On-device Robot Computing Chip: 3rd Generation Intel Core Ultra
  • 3rd Generation Intel Core Ultra: 18A Process
  • 3rd Generation Intel Core Ultra GPU Architecture: Xe3
  • 3rd Generation Intel Core Ultra Equipped with NPU 5: Optimized Specifically for AI Tasks
  • 5.9 MediaTek
  • Evolution and Future Development of EAI Chips
  • Genio Pro, Genio 420, Genio 360
  • Dimensity 9400, Dimensity 9400+
  • Genio Pro
  • Genio 420
  • Genio 360
  • Support MediaTek NeuroPilot AI Software Development Kit
  • 5.10 Li Auto
  • Parameters of Mach M100
  • Self-Developed Chip Mach M100
  • Self-Developed Chip Mach M100: Internal Structure
  • Self-Developed Chip Mach M100: CPU Structure
  • Self-Developed Chip Mach M100: NPU Structure
  • 5.11 HOUMO.AI
  • Evolution and Future Development of Embodied Intelligence Chips
  • Houmo Manjie M50 Chip (1)
  • Houmo Manjie M50 Chip (2)
  • Houmo Manjie M50: Equipped with the "Tianxuan" Architecture - Self-developed Second-Generation Compute-in-Memory IPU Design
  • Houmo Manjie M50 Toolchain: Houmo Dadao
  • 5.12 Huixi Intelligent Technology
  • Evolution and Future Development of Embodied Intelligence Chips
  • Huixi R1 (1)
  • Huixi R1 (2)
  • Self-developed Turing-Complete Instruction Set
  • Self-developed RPU Neural Network Accelerator
  • Innovative Functional Safety Architecture RIF