如果把特斯拉电动车的核心供应商名单,与Optimus人形机器人的潜在供应商名单放在一起,会看到一个耐人寻味的现象:那些为特斯拉提供底盘、热管理的厂商,如今又齐齐出现在Optimus执行器、关节模组和传感器的候选供应链中。这不是巧合,而是汽车供应链能力外溢的一种自然延伸。盖世合伙人、研究院副总裁王显斌的判断是,汽车与机器人目前的供应链重合度大约在60%到70%。
If you put Tesla's core supplier list for electric vehicles side by side with the potential supplier list for the Optimus humanoid robot, you will see an intriguing phenomenon: those manufacturers that supply Tesla with chassis and thermal management are now appearing together in the candidate supply chains for Optimus actuators, joint modules, and sensors. This is not a coincidence, but a natural extension of the spillover of automotive supply chain capabilities. Wang Xianbin, Gasgoo partner and vice president of the research institute, judges that the current supply chain overlap between automobiles and robots is about 60% to 70%.

六七成同源同的是什么
同的是产业能力,不是现成产品。概括起来,主要集中在三个方面:核心零部件可迁移、制造能力可复用、车规标准可借鉴。目前业内普遍认为,具身机器人与智能汽车遵循同一套感知、决策、执行架构。感知层,摄像头、激光雷达是共同的眼睛;决策层,高算力芯片和域控制器是共同的大脑硬件,人形机器人也可以在底层软件架构层面复用智能驾驶技术;执行层,电机、减速器是共同的四肢;电池、热管理这些能源与散热系统,则是共同的血液和循环系统。正是这种架构相通,在汽车产业链上催生了明显的一对多模式。用车百会研究院理事长张永伟的话来说,一个部件既可以支持汽车,也可以支持具身机器人。
What exactly is the 60% to 70% shared origin
What is shared is industrial capability, not ready-made products. To summarize, it is mainly concentrated in three aspects: core components can be migrated, manufacturing capabilities can be reused, and automotive-grade standards can be referenced. At present, the industry generally believes that embodied robots and intelligent vehicles follow the same set of perception, decision-making, and execution architecture. At the perception layer, cameras and lidar are the shared eyes; at the decision-making layer, high-computing-power chips and domain controllers are the shared brain hardware, and humanoid robots can also reuse intelligent driving technology at the underlying software architecture level; at the execution layer, motors and reducers are the shared limbs; and energy and cooling systems such as batteries and thermal management are the shared blood and circulatory system. It is precisely this architectural commonality that has given rise to an obvious one-to-many model in the automotive industry chain. In the words of Zhang Yongwei, chairman of the China EV100 Research Institute, one component can support both automobiles and embodied robots.
越便宜越同源,越贵越不同源
剩下30%到40%的不同源,是汽车和具身机器人各自所特有的部分,比如灵巧手、高自由度关节模组、力触觉传感器、微型电机、小脑运控算法、具身大脑等。这部分恰恰也是整机最贵的地方。以执行器为例,整体约占整机成本的40%到60%。虽然电机、结构件和总成制造可以依托汽车供应链,但执行器内部价值最高的高精度减速器、力矩传感器和一体化集成,并不由传统汽车零部件体系独立覆盖,仍需机器人领域的专业供应商参与。换句话说,近七成的产业链重合度固然乐观,但真相是:越便宜、越标准化的环节越同源,越贵、越决定性能的环节越不同源。
The cheaper it is, the more shared the origin; the more expensive it is, the less shared the origin
The remaining 30% to 40% that is not shared is the part unique to automobiles and embodied robots respectively, such as dexterous hands, high-degree-of-freedom joint modules, force-tactile sensors, micro motors, cerebellum motion control algorithms, and embodied brains. This part is precisely also the most expensive part of the complete machine. Taking actuators as an example, they account for about 40% to 60% of the complete machine cost overall. Although motors, structural parts, and assembly manufacturing can rely on the automotive supply chain, the highest-value high-precision reducers, torque sensors, and integrated integration within actuators are not independently covered by the traditional automotive parts system and still require the participation of specialized suppliers in the robotics field. In other words, a nearly 70% industrial chain overlap is certainly optimistic, but the truth is: the cheaper and more standardized the link, the more shared the origin; the more expensive and more performance-determining the link, the less shared the origin.
同源不等于照搬
智能汽车和机器人的AI范式高度同源,但历史不会简单重复。东风汽车集团研发总院智能化技术首席总工程师张振林说,这背后两个领域最根本的分野藏在一个看似简单的差异里——应用场景不同。智能驾驶主要解决的是车辆如何在相对受限的物理空间中,通过环境感知、自主决策和避障,从A点到达B点。但机器人,尤其是面向真实作业的机器人,最大的不同是存在大量高频、主动的接触交互,其中涉及力、摩擦、动力学、形变等一系列问题。换言之,智能驾驶更多是在约束条件下学习如何穿行于世界,而具身智能机器人还必须进一步学习如何作用于世界,以及如何应对动作带来的结果和反馈。从二维到三维,多出来的那一维,全是难题。一个现实挑战是,具身机器人体积往往比汽车小得多,自由度却更高,这对控制器散热、结构设计和系统集成的要求,与汽车完全不在一个量级。
Shared origin does not mean copying
The AI paradigms of intelligent vehicles and robots are highly similar in origin, but history will not simply repeat itself. Zhang Zhenlin, chief chief engineer of intelligent technology at Dongfeng Motor Group's R&D General Institute, said that behind this, the most fundamental divide between the two fields is hidden in a seemingly simple difference—different application scenarios. Intelligent driving mainly solves how a vehicle, in a relatively constrained physical space, gets from point A to point B through environmental perception, autonomous decision-making, and obstacle avoidance. But robots, especially robots oriented toward real operations, differ most in that there are a large number of high-frequency, active contact interactions, involving a series of issues such as force, friction, dynamics, and deformation. In other words, intelligent driving is more about learning how to pass through the world under constrained conditions, while embodied intelligent robots must further learn how to act on the world, and how to respond to the results and feedback brought by actions. From two dimensions to three dimensions, the extra dimension is all difficult problems. One practical challenge is that embodied robots are often much smaller than automobiles but have higher degrees of freedom, which places requirements on controller heat dissipation, structural design, and system integration that are completely on a different order of magnitude from automobiles.
双向奔赴没那么简单
最突出的矛盾是供需错位。法雷奥中国CTO顾剑民提到一个现实问题:零部件企业开口就问我能供你多少产品,机器人公司第一句却是我能卖给你多少机器人,出发点完全不同。汽车产业链玩家最常被提及的一张牌是躺在数据金矿上,但张振林说得很直白:有场景,不等于有数据;有数据,也不等于有高价值数据。对汽车产业链企业而言,真正重要的其实不是有场景本身,而是能不能把场景转化为持续的数据生产能力。顾剑民的判断是,两到三年后,将可以看到机器人率先在很多柔性产线、高强度、高容错率的岗位上落地;五年后,才是完全通用机器人大规模应用的时间点。当原生公司陆续在资本市场和交付榜单上跑起来时,汽车企业手里的同源牌,更像一张入场券,而不是通行证。
Two-way pursuit is not that simple
The most prominent contradiction is the mismatch between supply and demand. Gu Jianmin, CTO of Valeo China, mentioned a real problem: parts companies open their mouths and ask, "How many products can I supply you?" while robot companies' first sentence is, "How many robots can I sell you?" The starting points are completely different. The card most often mentioned for players in the automotive industry chain is that they are sitting on a data gold mine, but Zhang Zhenlin said it very bluntly: having scenarios does not equal having data; having data does not equal having high-value data. For enterprises in the automotive industry chain, what really matters is not having scenarios themselves, but whether they can convert scenarios into continuous data production capability. Gu Jianmin's judgment is that in two to three years, we will see robots first implemented in many flexible production lines and in positions with high intensity and high fault tolerance; five years later will be the time point for large-scale application of fully general-purpose robots. When native companies successively get running in the capital market and on delivery rankings, the shared-origin card in the hands of automotive companies is more like an admission ticket than a pass.
