摩根士丹利在5月7日出了一份人型机器人研报<<Humanoid Horizons: Money Meets Machines4>>。研报认为：“大脑未定型，数据是卡点”。今晚我们就通过大摩的观点，了解一些人型机器人的现状。
On May 7, Morgan Stanley released a research report on humanoid robots <<Humanoid Horizons: Money Meets Machines>>. The report states: "The brain is not yet formed; data is the dead end." Tonight, through Morgan Stanley's perspective, we'll explore the current state of humanoid robots.
1、 按马斯克的观点，机器人可类比于汽车，约2年迭代，到第三代产品时，有望实现5至10倍的优化提升。因此Optimus V3的量产虽然一再被推迟，但市场还是对其潜在的技术突破充满期待。据大摩对特斯拉的跟踪，其第三代Optimus的设计有望在2026年中亮相，量产目标定于2026年7月下旬至8月。马斯克表示，由于Optimus是全新生产线的全新产品，量产初期的产能爬坡速度会相当缓慢。
According to Musk, robots can be compared to cars, iterating about every two years, and by the third generation, they are expected to achieve a 5 to 10-fold improvement in optimization. Therefore, although mass production of Optimus V3 has been repeatedly delayed, the market remains highly optimistic about its potential technological breakthroughs. According to Morgan Stanley's tracking of Tesla, the design of its third-generation Optimus is expected to debut in mid-2026, with mass production targeted for late July to August 2026. Musk stated that since Optimus is a brand-new product on a completely new production line, the ramp-up speed in the initial phase of mass production will be quite slow.
2、 大摩认为，到2036年，人形机器人的保有量将达到约2440万台。到2050年，人形机器人的累计保有量将达到约10亿台。假设更换周期为6年，并基于平均售价（ASP）假设，大摩预计，到2050年，全球人形机器人市场的年收入可达7.5万亿美元（到2035年为3220亿美元，到2040年为2.0万亿美元）。作为对比，2024年全球前20大汽车制造商的营收合计约为2.5万亿美元。
Morgan Stanley believes that by 2036, the number of humanoid robots will reach about 24.4 million units. By 2050, the cumulative number of humanoid robots will reach about 1 billion units. Assuming a replacement cycle of six years and based on average selling price (ASP), Morgan Stanley expects the global humanoid robot market to generate annual revenue of $7.5 trillion by 2050 ($322 billion by 2035 and $2.0 trillion by 2040). For comparison, the combined revenue of the world's top 20 automakers in 2024 is about $2.5 trillion.
3、 人形机器人的商业化正在加速，中国的系统集成商正从演示阶段迈向真实世界的试点应用，覆盖巡检、旅游、服务、制造、物流及家庭辅助等领域，同时训练方式也越来越多地向强化学习转变。从产量来看，越疆科技和智平方科技各有1000台机器人，优必选有5000台工业人形机器人。
The commercialization of humanoid robots is accelerating, with Chinese system integrators moving from the demonstration stage to real-world pilot applications, covering inspection, tourism, services, manufacturing, logistics, and home assistance, while training methods are increasingly shifting toward reinforcement learning. In terms of output, Yuejiang Technology and Zhipingfang Technology each have 1,000 robots, while UBTECH has 5,000 industrial humanoid robots.
4、 机器人大脑技术体系仍未定型，世界模型和VLA模型是主流。竞争壁垒正逐渐向拥有专有数据（特别是十分稀缺的真实世界机器人数据）的方向倾斜。而中国企业在人工智能、数据基础设施和硬件领域采用的全栈模式，有助于在行业实现模块化之前加快迭代速度并推动早期商业化。
The technology system for robot brains is still unsettled, with the world model and VLA model being mainstream. Competitive barriers are gradually tilting toward proprietary data, especially the extremely rare real-world robot data. Meanwhile, the full-stack model adopted by Chinese companies in AI, data infrastructure, and hardware helps accelerate iteration and drive early commercialization before the industry achieves modularization.
合上研报，回观国内，机器人智能化水平不仅需要数字数据，还要求物理世界里的各种接触，比如摩擦力、重力、黏滞力等这种高维数据。智元机器人正在探索一条新的数据采集路径，从需要人配合机器人操作，到人
佩戴轻量设备，再到无感的可穿戴产品，即在人的日常工作中采集积累各种数据。因此智元新发布了MEgo系列无本体采集硬件。
Closing the research report and looking back at China, the level of robot intelligence requires not only digital data but also various physical contacts, such as friction, gravity, and viscous force—high-dimensional data. Zhiyuan Robotics is exploring a new path for data collection, from requiring human assistance to operate robots, to wearing lightweight devices, and then to imperceptible wearable products—collecting and accumulating various data in daily human work. Therefore, Zhiyuan has newly released the MEgo series of bodyless collection hardware.
Optimus和Tesla一样，犹如鲶鱼，希望它能给我们带来惊喜，同时也推动行业竞争和发展。
Same as Tesla, Optimus is like a catfish; we hope it can surprise us while also driving industry competition and development.
