“# J.P.Morgan ## Global Humanoid Physical AI update on the automotive sector In this note, we summarise some of the latest developments in Physical AI with a focus on the global automotive sector including names such as BMW, Schaeffler, Toyota, Humanoid, Tesla and Wandercraft amon…”
关联内容单元 #16435 · 原始来源原始全文
# J.P.Morgan ## Global Humanoid Physical AI update on the automotive sector In this note, we summarise some of the latest developments in Physical AI with a focus on the global automotive sector including names such as BMW, Schaeffler, Toyota, Humanoid, Tesla and Wandercraft amongst other names. Tesla's Elon Musk described, in our view, very well the humanoid development as slow and “somewhat bumpy” because Optimus is a radically new product—comparing the process to an S-curve whose initial slope is shallow, with progress limited by the slowest of roughly 10,000 things that must go right. BMW – Capital Markets Update on Physical AI: Developing a “Physical AI” robotics stack centred on Plant Landshut, combining open software, production data, simulation, motion planning and demonstration-based training. Leipzig and Spartanburg pilots complement the effort, with emphasis on flexible, dexterous factory tasks and staged deployment. Mercedes-Benz: At the Mercedes-Benz Hungary plant opening, we met Apptronik CEO Jeff Cardenas, who outlined Apptronik's strategy and its partnership with Mercedes-Benz, positioning the company to support automation across multiple shopfloor workflows. A key takeaway was that humanoids need not be strictly two-legged, wheeled platforms and humanoid arms were also cited for tasks such as box handling, parts handling, body-shop loading and bin picking. Mercedes-Benz sees assembly-line deployment becoming realistic in the 2030–2035 timeframe as dexterity and platform maturity improve, with Apptronik emphasizing safety and a task-fit approach matching mobility/dexterity to each application. Schaeffler: Pursuing humanoids as a component supplier, industrial user and integration partner. The strategy combines humanoid training facilities, factory data, actuator and strain-wave gearbox content, automotive-grade cost reduction and partnerships; motion components could represent roughly half of a humanoid bill-of-materials value. The firm has opened a state-of-the-art lighthouse factory for humanoid robotics in Taicang, China. The Lighthouse Factory marks the start of mass production of these components for Schaeffler and offers customers and partners the opportunity to train and test physical AI systems in actual industrial conditions. The company recently announced the acquisition of two new customers from the Greater China region in Q3 2026. With this, global incoming orders for humanoid robotics rose to some 400 million euros. Humanoid: Advancing wheeled humanoids for industrial, automotive, logistics and service workflows. The n ecosystem spans Bosch for manufacturing, Schaeffler for actuators and deployment, Siemens for industrial validation, and NVIDIA for simulation, compute and reinforcement-learning infrastructure. Please follow the link for feedback on our recent tour of their Humanoid robots. Tesla: Tesla (covered by Rajat Gupta) has begun early production of production-quality Optimus humanoid robots on high-volume lines at its Fremont, California factory (in space previously used for vehicles), as announced in its Q3 2026 See page 13 for analyst certification and important disclosures, including non-US analyst disclosures. European Autos & Auto Parts Jose M Asumendi AC (44-20) 7742-5315 jose.m.asumendi@jpmorgan.com J.P. Morgan Securities plc Piyush Singla (91-22) 6157-3324 piyush.singla@jpmorgan.com J.P. Morgan India Private Limited Specialist Sales contact details: Sam Edmunds - Specialist Sales - European Industrials (44-20) 7742-8733 sam.edmunds@jpmorgan.com shareholder update. Recent comments from Elon Musk during the latest third quarter results: Musk described this as “early production.” He emphasized that the ramp will be slow and “somewhat bumpy” because Optimus is a radically new product—comparing the process to an S-curve whose initial slope is shallow, with progress limited by the slowest of roughly 10,000 things that must go right. The presentation of the Tesla Roadster was probably one of the best car product launches ever seen, breaking all the physics rules in the world of road cars, showcasing the technology breadth of the firm when partnering with Space X. Neura Robotics: NEURA Robotics and TUM opened a 2,300+ sqm TUM RoboGym powered by NEURA near Munich Airport, designed to combine research, robot training and industrial task validation, while launching its next-gen humanoid 4NE1 Gen 4. NEURA also introduced Neuron AI, its in-house multimodal Physical-AI foundation model, intended to power deployments across five robot embodiments and enable faster customer task fine-tuning via Neuraverse. Separately, NEURA partnered with NXAI to integrate xLSTM for long-horizon memory to support multi-step task continuity, and formed a dedicated HealthTech unit—highlighting an autonomous bed-transport robot with an estimated \~1,300 hours/year time savings for care teams. Toyota: Toyota (covered by Akira Kishimoto) plans large-scale factory automation, including AI-enabled humanoids trained from veteran-worker demonstrations. Toyota, which aims to share learned skills globally, is developing diffusion-based Large Behavior Models, and is also testing Agility Robotics' Digit for logistics applications. Toyota has provided one of the largest commitments to Humanoids. Toyota Group announced plans to invest \$6.42 billion annually from 2028 to update factories and deploy 400,000 robots worldwide, including AI-powered humanoid workers. These robots will learn the skills of veteran workers through cameras and sensors, enabling skill transfer without complex programming. The deployment covers 150,000 robots at Toyota Motor Corp. and 250,000 at group companies like Denso and Aisin. Honda: Honda (covered by Akira Kishimoto) highlighted a renewed push into humanoid/teleoperated robotics at the 2026 Japan Mobility Show in Tokyo, demonstrating its wheel-mounted Avatar robot equipped with an AI-enhanced, four-finger mechanical hand capable of delicate manipulation. Honda CEO Toshihiro Mibe said the company is already rethinking future vehicle design from the earliest development stages to make cars easier to assemble with humanoid robot workers, drawing a parallel to prior design changes driven by earlier shifts in assembly-line automation. Honda also noted it is testing the Avatar Robot on factory tasks in Japan—such as wire-harness installation—positioning avatar robotics and hand technologies as potential future growth businesses. Mitsubishi: Mitsubishi (covered by Akira Kishimoto) said in July that it will partner with Japanese robotics startup Highlanders Inc. to develop humanoid robots intended for deployment in its factories, and plans to mass-produce the robots at its engine plant as early as 2027—positioning the initiative as part of its efforts to expand beyond its core automotive business. Xpeng at the Paris Autoshow provided an update on Physical AI : XPeng (covered by Nick Lai) is moving its humanoid robot, IRON, toward n, with mass production targeted for late 2026, initial deliveries expected in 1H27, and a European rollout planned for 2H27. On Sept'26, the company commissioned what it described as the world's first automated production line for advanced humanoid robots, with more than 80% of core manufacturing processes automated; notably, the first IRON unit reportedly autonomously walked off the line at completion.  ## BMW During the latest Capital Markets Day, BMW outlined its "Physical AI" strategy. BMW Group's work in physical AI, particularly the role of BMW Group Plant Landshut in developing AI-supported robotics for component production. Landshut is taking on central software development tasks covering open software platforms, training data, simulation, motion planning, and the training of robotic systems. This work complements the pilot projects exploring humanoid robotics at Plants Leipzig and Spartanburg. What makes the Landshut project especially relevant is the opportunity to develop and evaluate these technologies in real production environments. The plant's broad range of products and manufacturing processes provides a practical setting for assessing where humanoid robotic systems could add value, particularly in tasks requiring flexibility, fine motor skills, and an understanding of the surrounding work environment. The team is developing software that enables robots to perceive their surroundings, assess situations, and determine appropriate actions. Figure 1: Physical AI PHYSICAL AI UNLOCKS IMMEDIATE VALUE BY AUGMENTING ROBUST AUTOMATION SYSTEMS WITH AI-DRIVEN INTELLIGENCE. Figure 2: Physical AI  INDUSTRIAL HUMANoids WITH BMW DATA & TRAINING CAN BE APPLIED TO A BROAD RANGE OF PRODUCTION AND LOGISTICS TASKS. - Built on proven and robust industrial hardware. - AI enhances flexibility (perception, adaptability and decision-making) and thereby the application scope. - Focused on a limited set of high-value production tasks. - Enables quick wins with immediate business impact. Source: Company data.  Source: Company data.  Figure 3: Physical AI GENERAL-PURPOSE PHYSICAL AI ENABLES A SCALABLE WORKFORCE ACROSS INDUSTRIES AND ENVIRONMENTS.  GENERAL PURPOSE ■ Human-like mobility and manipulation capabilities ■ Advanced whole-body control for complex tasks Source: Company data.  The project also focuses on creating a flexible robotics ecosystem. Open software platforms combine traditional, rule-based programming with modern Vision-Language-Action models, which connect visual information and language understanding with robotic movement. Reliable data is then collected from production environments and test setups and used in simulations to model movement patterns and motion planning. This allows new processes to be tested before they are introduced into ongoing production and supports a gradual progression from pilot projects to wider applications. In parallel, humanoid robotic systems learn new skills through demonstrations. Project teams record movement patterns and work sequences using camera systems, motion-capture suits, and data gloves. The resulting information is converted into generalisable behavioral models, known as policies, which can support the development of more adaptable robotic systems. This work matters because it connects advanced software development with practical manufacturing experience. By testing these technologies carefully and scaling them in stages, the BMW Group can better evaluate their potential benefits for production. ## Mercedes-Benz During the Mercedes-Benz Hungary plant opening, we met with Apptronik CEO Jeff Cardenas, who outlined Apptronik's strategy and its partnership with Mercedes-Benz, and highlighted the core strengths the company believes position it to deliver automation solutions for Mercedes-Benz across a range of applications. It is important to note that the implementation of Humanoids does not need to be in the form of two legged versions; it can also adopt different forms including wheels or Humanoid arms, across a range across different applications including box handling, cast parts handling, body shop part loading, bin picking amongst other actions. Mercedes Benz sees the implementation of Humanoids on the assembly line becoming a reality by 2030 -2035. By then, Humanoid Dexterity and Humanoid development should have considerably evolved. Jeff Cardenas, the CEO of Apptronik, also mentioned several end markets that could become meaningful going forward defined as a) Back of the house: manufacturing and logistics; b) Front of the house: healthcare, retail and shops; and c) Home applications: applications in the home and home care, this latter could represent the largest opportunity for the firm. Having said this, Apptronik reiterated the concept of safety when implementing Humanoids in the manufacturing environment as well as the business model of offering a robotic solution for the industrial environment correctly adapted to the level of dexterity and mobility required for the task. Figure 4: Mercedes-Benz is testing the use of Apollo, a humanoid robot from Apptronik, in production.  Source: Mercedes Benz ## Schaeffler From Schaeffler's Sep'26 Humanoid Tour, we understand the company has outlined a relatively concrete path to participate across the humanoid stack, which is anchored on motion-component content, factory deployment learnings/data capture, and ecosystem partnerships, rather than positioning humanoids as a purely software-led opportunity. 1. Physical AI: Why Schaeffler is Leaning into Humanoids - Schaeffler believes humanoid adoption to be a combination of i) AI moving from machine learning → generative AI → agentic AI → physical AI, and ii) macro labor tightness (\~85m unfilled positions globally and \~800m industrial workers) as the economic driver for automation to expand beyond classic industrial robotics footprints as they view 60% of tasks as automatable. 2. Early Adoption: Humanoid Gyms + Factory Footprint - Schaeffler positions early adoption as a compounding advantage as they can deploy, collect data, train skills, and industrialize against the backdrop of their own manufacturing base. Specifically, the company highlights regional humanoid gyms as infrastructure for data collection and manufacturing-skill development, including what it calls the world's first combined Humanoid Gym & humanoid actuator production facility in Shanghai, with a grand opening planned for Q4 2026. The stated training loop spans robot/teleoperation data, synthetic data and peripheral data, feeding simulation and robot foundation models (examples for Nvidia / Gr00t) and then deployment across Schaeffler use cases (logistics, feeding/tending, assembly, inspection, packaging). 3. System Understanding: Use-Case-Driven Engineering - A notable element is the use-case-driven development method: Schaeffler shows an engineering loop that starts with use-case torque-cycle analysis and requirements (thermal design, cycle efficiency, load cycles) and then drives actuator design choices, with simulation explicitly tied to tasks such as box handling, walking with payload, pick-and-place, and gripper manipulation. 4. Motion technology as the primary scaling lever - On the economics, Schaeffler views its integrated linear + rotary actuator competence accounts for \~50% of humanoid BOM costs. On productization, the company points to a new humanoid rotary actuator platform with shared controllers/e-motors across multiple sizes, \~20% reduced axial packaging and up to \~500g of weight reduction, with an integrated torque sensor targeted for Q4 2026; management also flags 30+ actuators per humanoid as the underlying content-per-unit opportunity. Cost-down is positioned as a function of leveraging automotive-grade processes and vertical integration, including \~30% manufacturing cost reduction claims across selected components/process routes (e.g., strainwave gearbox and bearing-related manufacturing), supported by a track record of \~2 million Electronic Cam Phasers (ECPs) sold since 2015. 5. Partnerships as a go-to-market accelerator - Schaeffler positions itself across multiple roles in the ecosystem i.e. humanoid user, system & component supplier (broad motion portfolio + joint development), integrator/ n partner (turn pilots into production with line-ready processes), and lifecycle solution provider. Company also lists partnership targets spanning OEMs/robotics players (e.g., Neura, Agility, Hexagon, and others) as part of the market-access pathway. Figure 5: Overview of Schaeffler Product Fields  Work in progress – conception phase Source: Company data. Figure 6: Building Strong OEM Strategic Partnerships  Source: Company data. Figure 7: Physical AI: Future of Manufacturing  Source: Company data. ## Latest News & Announcements \- Hexagon Robotics and Schaeffler announced that the AEON humanoid has entered Schaeffler's Humanoid Gym in Germany, marking a new milestone toward the planned deployment of at least 1,000 AEON units in the coming years. The Humanoid Gym uses a Train-Validate-Deploy model in a dedicated industrial environment, allowing the partners to teach AEON new industrial capabilities while building Schaeffler's in-house expertise to operate, train, and integrate humanoids into production. The effort is intended to prepare both the robots and the manufacturing ecosystem for deployment and expanded factory-floor use cases. \- Schaeffler is expanding its humanoid robotics product portfolio and bringing formed strain wave gearboxes to market. Strain wave gearboxes are key actuator components that can account for roughly half of a humanoid robot's manufacturing cost. The gearboxes enable precise motion transmission with high torque capacity by delivering high gear reduction in a compact form factor. \- Schaeffler has opened a state-of-the-art lighthouse factory for humanoid robotics in Taicang, China. It houses an intelligent assembly line, a training and data acquisition center, a platform for digital twins, and a data center across approximately 7,500 square meters of floor space. The building covers the entire process from testing of individual robot skills to application in the actual manufacturing environment, showing what “humanoids making humanoids” looks like in practice: humanoid robots assembling key joint modules – including actuators – for the next generation of humanoid robots. The Lighthouse Factory marks the start of mass production of these components for Schaeffler and offers customers and partners the opportunity to train and test physical AI systems in actual industrial conditions. It therefore serves to drive adoption of humanoids in industry. Schaeffler’s business also reflects the booming demand in this growth field: The company recently announced the acquisition of two new customers from the Greater China region in Q3 2026. With this, global incoming orders for humanoid robotics rose to some 400 million euros. ## Humanoid In September, we met with Humanoid's management team, toured its facilities, and discussed the company's medium-term technical roadmap. We came away positively surprised by the pace of progress and observed wheeled humanoids executing picking and sorting workflows that appear relevant for industrial/automotive environments as well as logistics and services. Strategically, Humanoid is building an ecosystem that, in our view, directly targets the key de-risking steps required for scalable European manufacturing and plus Design-for-Excellence support. Schaeffler brings both actuator supply/co-development and the potential for meaningful end-customer deployment at scale, creating a two-way loop where factory data can feed product iteration and AI skill development. Siemens serves as an industrial validation venue, where PoC logistics loops help establish reliability and integration credentials. Finally, NVIDIA underpins the AI/simulation stack, supporting simulation (Isaac Sim/Omniverse), accelerated compute/edge processing, and reinforcement-learning-based skill training. ## Latest News & Announcements \- Humanoid is expanding its collaboration with NVIDIA to underpin the onboard compute and runtime architecture f…
AI 解读
本次更新披露了多家汽车及零部件企业从机器人试点、早期生产向量产准备推进的具体进展，包括宝马的Physical AI技术栈、梅赛德斯-奔驰与Apptronik的Apollo测试、舍弗勒约4亿欧元新增订单、特斯拉Optimus早期生产、丰田自2028年起的机器人投资、小鹏IRON量产及交付时间表，以及NEURA的训练和应用场景。