文 |财观二姐 可以说2025年人形机器人赛道在全球升温，离不开的一个推手就是马斯克和他的特斯拉----从2021年定 名为Optimus（擎天柱）的人形机器概念图公开、到2022与2023年分别公开原型机与第二代（Gen 2）产 品，三年里每年一个脚印；2024年Optimus未再推出新版本，只是进行了部分功能场景展示，从而让业 界的胃口被吊到了2025年。 2025年3月，马斯克就在特斯拉的全员大会上宣布要在年内生产出5000台Optimus，但仅仅4个月后该款 机器人产能不足的信息便在财报电话会议上被确认，宣告了Optimus进展目标在年内的跳票。 考虑到马斯克有过几次的"画饼"前科，被他寄于厚望的这款人形机器人以及其量产前景会不会再出波 折，就很难讲了。 也许人们会好奇，这位"当代钢铁侠"所领军的特斯拉----这家当今全球可以说资源技术禀赋最为全面、 同时也是最擅于不断用一个个成功案例来推动其公司叙事的企业，为什么会连续两年没能在他们下重注 的人形机器人赛道上取得大的进展？这背后，对于国内目前势头火热的人形机器人领域玩家又提供了什 么样的启示？ 场景之别 人形机器人的场景差异，可能是以特斯拉Optim ...

Core Viewpoint - Tesla is shifting to a single, large end-to-end neural network that directly generates control actions from pixel and sensor data, eliminating explicit perception modules [1][34]. Group 1: Reasons for Transition to End-to-End Neural Networks - Integrating human values (like driving smoothness and risk assessment) into code is extremely challenging [3]. - Poor interface definitions between traditional perception, prediction, and planning can lead to information loss [4]. - The end-to-end approach is easier to scale for handling long-tail problems in the real world [5]. - It allows for homogeneous computation with deterministic latency, which is crucial for real-time systems [6]. Group 2: Challenges in Learning "Pixel to Control" - The primary challenges include the curse of dimensionality, interpretability and safety guarantees, and evaluation [7][8][9]. - The input context can be extensive, with a 30-second window potentially reaching 2 billion tokens [10][49]. - Tesla leverages its vast fleet data to extract valuable corner case data through complex, trigger-based data collection methods [11][51][56]. Group 3: Solutions to Challenges - For the curse of dimensionality, Tesla refines its extensive driving data to ensure the right correlations are captured [51][56]. - Interpretability is addressed by prompting the end-to-end model to predict various auxiliary outputs for debugging and safety assurance [12][60]. - Evaluation challenges are tackled by creating a neural network-based world simulator that can generate consistent video streams from multiple cameras [19][79]. Group 4: Future Developments - The next step involves the Cyber Cab, a next-generation vehicle designed specifically for robotaxi services, utilizing the same neural network technology [25][83]. - The technology developed for autonomous driving is also being adapted for humanoid robots, such as Optimus [26][86].

Core Insights - The competition between Elon Musk's Tesla and Brett Adcock's Figure in the humanoid robot sector highlights two distinct approaches to realizing the future of robotics [4][28] - Figure's recent product launch, Figure 03, has gained significant public attention, contrasting with Tesla's more reserved approach to showcasing its Optimus robot [2][28] Company Overview - Tesla and Figure entered the humanoid robot market around the same time, with Musk announcing the Tesla Bot (later named Optimus) in August 2021 and Adcock founding Figure AI shortly thereafter [5][6] - Both companies have made significant strides in humanoid robot development, with Tesla focusing on engineering and system building, while Figure emphasizes public engagement and practical demonstrations [4][24] Product Development Timeline - Tesla's Optimus has seen several iterations, including the Bumble C prototype in 2022 and the Gen1 and Gen2 models in 2023, with a focus on refining its capabilities [6][7] - Figure has rapidly progressed with its robots, launching Figure 01 in March 2023 and Figure 02 in August 2024, showcasing practical applications in everyday settings [6][9] Public Perception and Marketing - Figure's marketing strategy involves demonstrating robots in relatable home environments, creating a tangible sense of how robots can integrate into daily life, while Tesla's focus remains on technical specifications and engineering challenges [3][4] - The contrasting narratives have led to Figure capturing more public attention, while Tesla's Optimus is perceived as a more internal project lacking public visibility [3][14] Future Projections - Musk envisions a future where humanoid robots could outnumber humans significantly, with production targets suggesting Tesla could manufacture up to 100 million units annually [15][28] - However, challenges in production, supply chain issues, and team instability have led to delays in Tesla's robot rollout, raising questions about its ability to meet these ambitious goals [19][26] Strategic Focus - Figure's singular focus on humanoid robots allows it to allocate all resources towards this goal, while Tesla's broader responsibilities as an automotive and energy company may hinder its agility in the robotics space [20][25] - The competition reflects a generational shift, with Adcock embodying a more agile, startup mentality compared to Musk's established corporate approach [27][28]

Core Insights - The Chinese humanoid robot industry is rapidly advancing, leveraging manufacturing advantages and significantly reducing costs, making robots more accessible to consumers [1][2][4] - While China excels in hardware production and cost control, the U.S. maintains an edge in software ecosystems and AI capabilities, particularly in developing intelligent robots that can understand and interact with their environment [10][11][59] - The competition between Chinese and American humanoid robots is intensifying, with both sides focusing on different aspects of development: China on market penetration and cost reduction, and the U.S. on advanced AI and software integration [13][15][60] Industry Overview - The humanoid robot market is projected to experience explosive growth, with estimates suggesting that the global market could reach 1.1 trillion yuan by 2035, and the Chinese market alone could achieve 300 billion yuan [23][24] - As of mid-2025, over 220 humanoid robot companies exist globally, with Chinese firms accounting for more than half of this total [27] - The Chinese humanoid robot sector is witnessing a surge in enterprise registrations, with over 105 new companies established in the first half of 2025, reflecting a significant shift towards commercialization [42] Technological Developments - Chinese companies are focusing on specific industrial applications for humanoid robots, such as quality inspection in automotive manufacturing and precision tasks in agriculture [5][6][60] - Despite advancements, Chinese humanoid robots still face challenges in basic capabilities like motion control and autonomy, indicating a need for further technological development [31][34] - The U.S. is making strides in creating humanoid robots with advanced AI capabilities, such as Tesla's Optimus, which is designed to perform complex tasks and adapt to various environments [38][50][55] Market Dynamics - The competition is characterized by a divergence in strategies: Chinese firms prioritize cost-effective production and market capture, while American firms emphasize software innovation and AI integration [15][62] - Significant investments are flowing into the humanoid robot sector, with over 14 billion yuan raised globally in the first half of 2025, and Chinese companies securing a substantial portion of this funding [24][66] - The regional concentration of humanoid robot companies is notable, with the Yangtze River Delta region housing a significant share of enterprises, indicating a trend towards industrial clustering [43] Future Outlook - The humanoid robot industry is at a critical juncture, transitioning from basic mobility to functional task execution, with the potential for widespread application in various sectors due to labor shortages in aging societies [37][66] - The ultimate competition will hinge on the development of "embodied intelligence," which combines advanced AI with humanoid robotics, determining which country can produce robots that not only move but also think and adapt [19][64]

Core Viewpoint - Despite significant investments from venture capital firms and large tech companies, humanoid robots still struggle to achieve dexterity, which is essential for performing tasks in human environments [2][3][4]. Group 1: Historical Context of Humanoid Robots - The concept of humanoid robots has been explored for over 65 years, with early developments including a computer-controlled robotic arm capable of stacking blocks in 1961 [3]. - The evolution of humanoid robots has seen contributions from various institutions, including WABOT-1 from Waseda University in the 1970s and Honda's ASIMO in 2000 [11][12]. Group 2: Current State and Future Predictions - Humanoid robots are currently in the early stages of development, with Gartner indicating they have not yet reached their peak hype [4]. - Companies like Tesla and Figure are optimistic about the economic potential of humanoid robots, with predictions of creating trillions in revenue [9][10]. Group 3: Challenges in Dexterity - Achieving human-level dexterity in humanoid robots remains a significant challenge, as current robotic hands lack the necessary finesse and adaptability for a wide range of tasks [23][24]. - Existing methods for training robots often rely on visual demonstrations, which do not adequately capture the tactile feedback necessary for dexterous manipulation [27][28]. Group 4: Learning Approaches - The industry has seen a shift towards end-to-end learning methods, where robots learn from observing human actions, but this approach has limitations due to the lack of tactile feedback and precision [30][31]. - Successful applications of end-to-end learning in other fields, such as speech recognition and image labeling, highlight the importance of pre-processing and human-like structures in achieving effective learning outcomes [49][50]. Group 5: Importance of Tactile Feedback - Human dexterity is heavily reliant on rich tactile feedback, which current humanoid robots do not possess, leading to challenges in replicating human-like manipulation [51][52]. - The complexity of human touch perception and the integration of multiple body parts in dexterous tasks further complicate the development of humanoid robots capable of similar actions [52].

Core Viewpoint - Elon Musk emphasized that Tesla is fully committed to scaling the Optimus project, defining it as the most important product in the company's history, with expectations that it will account for 80% of the company's future value [2][6]. Group 1: Production Timeline and Goals - Tesla aims for internal limited production and testing of thousands of Optimus units by 2025, ramping up to 50,000-100,000 units for external sales in 2026, and targeting an annual production of 1 million units within five years [2]. - The current supply chain for Optimus is based on the design of Optimus V2, with actuators and sensors each accounting for approximately 30% of material costs [4]. Group 2: Supply Chain and Component Suppliers - The supply chain for Optimus includes Tier 1 suppliers such as Sanhua Intelligent Controls and Top Group for actuators, and Mingzhi Electric and Zhaowei Electromechanical for dexterous hands [7]. - Key component suppliers include Shuanghuan Transmission and Lide Harmony for reducers, Best for lead screws, and Rongtai Health for insulation parts [7]. - International Tier 1 suppliers include Amphenol for cables, TE Connectivity for six-dimensional torque sensors, and THK for lead screws [8]. Group 3: Challenges and Development Needs - Current challenges for Optimus include hardware issues such as overload and overheating of joint motors, insufficient dexterity and load capacity of dexterous hands, and the lifespan of transmission components [10]. - There is a need for improved compatibility between hardware and software, particularly in complex dynamic environments and multi-task coordination [10]. Group 4: Thermal Management Solutions - The thermal management system for Optimus V3 is similar to that of electric vehicles, focusing on the management of key components like batteries and motors [11]. - Sanhua Intelligent Controls is developing liquid cooling modules for Optimus, leveraging its experience in electric vehicle thermal management to address overheating issues in robotic joints [13]. - Sanhua plans to deliver approximately 2,000 actuators to Tesla by Q3 2025, with an annual order forecast of 5,000-10,000 units [13]. Group 5: Industry Trends and Future Outlook - Domestic suppliers are increasing investments to meet the demand for high-performance, miniaturized thermal management components for robots [14]. - The year 2025 is anticipated to be a pivotal year for humanoid robot mass production, with 2026 expected to be a critical turning point for the industry landscape [14]. - The upcoming iTherM 2025 conference will address advanced thermal management technologies and materials relevant to humanoid robots [15].

编辑丨 量子位 点击下方 卡片 ，关注" 具身智能之心 "公众号 >> 点击进入→ 具身 智能之心 技术交流群 更多干货，欢迎加入国内首个具身智能全栈学习社区 ： 具身智能之心知识星球 (戳我) ， 这里包含所有你想要的。 马斯克在忙着裁人，小扎这边继续忙着挖人。 这不？ Optimus AI团队负责人Ashish Kumar 决定离开特斯拉，加入Meta担任研究科学家。 至于离职感言，他表示： 带领Optimus AI团队的经历非常精彩和难忘。 我们全力推进可扩展方法——用强化学习取代传统技术栈，并通过视频学习来提升机器人的灵巧度。 他还进一步强调， 人工智能才是解锁人形机器人的最关键因素 。 与此同时，小扎砸钱挖人的形象已经深入人心，使得网友不禁锐评，有10亿美元吗？ Optimus团队负责人接连出走 那这位Optimus AI团队负责人到底是何大神？ Ashish Kumar ，UC伯克利博士，导师是被李飞飞尊称为"学术祖父"的Jitendra Malik教授，因在CV领域的研究而出名。 2015年，Ashish本科毕业于印度理工学院焦特布尔分校，之后到微软位于印度的实验室做了两年研究员，研究方向是资源 ...

Group 1 - Liu Qiangdong emphasized that private enterprises should not turn into enemies and that normal business competition should not become personal grudges [1][2] - Liu expressed respect for competitors and highlighted the importance of competition based on strategy, business models, and value creation [2] - JD.com is set to announce a new model for the hotel industry by the end of the year, aiming to optimize supply chain costs and avoid price wars [8] Group 2 - ByteDance announced it will proceed with TikTok-related work in accordance with Chinese laws, ensuring continued service for American users [2][3] - The Chinese government supports businesses in achieving solutions that comply with market rules and hopes for a fair operating environment for Chinese companies in the U.S. [3] Group 3 - The iPhone 17 series launched on September 19, with reports of scalpers marking up prices significantly, indicating strong demand [4] - The first-day sales performance of the iPhone 17 series suggests potential for better sales compared to the previous generation [4] Group 4 - Figure completed a Series C funding round, raising over $1 billion with a post-money valuation of $39 billion, aimed at expanding its AI platform and humanoid robots [5] - The participation of major industry players like NVIDIA and Intel highlights the importance of training power and data resources in future competition [5] Group 5 - Meta launched its first AI glasses with display capabilities, which can translate in real-time and support 3K video recording, marking a significant advancement in consumer wearable technology [11] - The global market for AI smart glasses is expected to grow significantly, with Meta projected to capture over 65% market share following the product launch [11] Group 6 - Seven Bulls Cloud and Wuxiang Cloud Valley formed a strategic partnership to scale AI inference computing power, targeting the burgeoning AI inference market [10] - The demand for AI computing power is expected to surge, with a shift towards rental models due to high costs and technical barriers [10]

据悉，在特斯拉任职期间，Ashish Kumar主导了Optimus AI团队的核心技术研发工作，其团队专注于通 过人工智能技术突破人形机器人的实用化瓶颈。他在社交平台发文中特别提到，团队"全力推进可扩展 方法——用强化学习取代传统技术栈，并通过视频学习来提升机器人的灵巧度"。 来源：环球网 强化学习作为人工智能领域的前沿技术，允许机器人通过试错自主优化行为策略，而非依赖预设程序。 Ashish Kumar团队此前展示的Optimus原型机已具备分拣电池、搬运物品等基础任务能力，其流畅的动 作控制被业界视为强化学习技术落地的标杆案例。此外，该团队通过视频学习技术，使机器人能够从人 类操作视频中提取动作模式，显著缩短了技能训练周期。 【环球网科技综合报道】9月20日消息，据多家外媒报道，特斯拉Optimus（擎天柱）人形机器人项目AI 团队负责人阿希什·库马尔（Ashish Kumar）已正式辞去在特斯拉的职务，并将于近期加入Meta（原 Facebook）公司担任研究科学家一职。当地时间9月19日，Ashish Kumar在个人社交平台发布长文，回 顾其在特斯拉的职业生涯，并透露了关于人形机器人技术发展的关键 ...

时令 发自 凹非寺 量子位 | 公众号 QbitAI 马斯克在忙着裁人，小扎这边继续忙着挖人。 这不？ Optimus AI团队负责人Ashish Kumar 决定离开特斯拉，加入Meta担任研究科学家。 至于离职感言，他表示： 带领Optimus AI团队的经历非常精彩和难忘。 我们全力推进可扩展方法——用强化学习取代传统技术栈，并通过视频学习来提升机器人的灵巧度。 他还进一步强调， 人工智能才是解锁人形机器人的最关键因素 。 与此同时，小扎砸钱挖人的形象已经深入人心，使得网友不禁锐评，有10亿美元吗？ Optimus团队负责人接连出走 那这位Optimus AI团队负责人到底是何大神？ Ashish Kumar ，UC伯克利博士，导师是被李飞飞尊称为"学术祖父"的Jitendra Malik教授，因在CV领域的研究而出名。 2015年，Ashish本科毕业于印度理工学院焦特布尔分校，之后到微软位于印度的实验室做了两年研究员，研究方向是资源高效的机器学习算 法。 2017年，Ashish从职场重返校园，开始到UC伯克利攻读博士，2023年7月他以ML科学家的身份加入特斯拉，一年多之后便成为擎天柱的AI 负责 ...