Q1
可以简单介绍一下您今天的演讲内容吗？
Could you briefly introduce your presentation topic today?
蒋峰 :
本次我演讲的内容是AI时代下的EE架构的一个发展演进。为什么选择这个话题呢？因为现在AI已经深度进入到我们工作中，不管生活中还是工作中都离不开AI，AI也在逐渐进入到我们汽车里面的一些功能体验。
所以说选择这个话题来跟大家分享一下。分享的主要是从以下几个方面来进行讲解：一个是回顾一下我们整个EE架构的发展历程，从传统的分布式到中央架构，再到现在我们面临AI即将发生的转变。
第二板块主要是讲一下我们现在面临AI的时代下，架构需要面临的一些技术变革。第三个方面的话，就是讲一下我们那个EE架构在AI时代下，我们的整个开发流程和评价体系，他是怎么来看，我们是怎么做一个比较合理的架构来面向我们的这个AI。
另外第四方面就是，从我们整个产业链还有一些开发团队的组织变革，来应对AI下架构发生的变化。最后的话，就是展望一下我们的EE架构，它现在虽然是在汽车上进行应用，但最终的话它其实可以赋能到我们整个具身智能产品生态链，包括人形机器人、无人机、飞行汽车等等这些智能具身产品。
Mr. Jiang Feng :
The topic of my presentation today is the evolution of EE architecture in the AI era. Why did I choose this topic? Because AI has already penetrated deeply into our work and daily life. AI is also gradually being integrated into various invehicle features and user experiences.
So I chose this topic to share my insights. My talk will mainly cover the following aspects: first, I’ll walk through the full evolution path of EE architecture, from traditional distributed architecture to centralized architecture, and now to the AIdriven transformation we’re facing.
The second part focuses on the technical shifts that EE architecture needs to undergo in the AI era. Third, I’ll discuss how our development process and evaluation system should view and build a reasonable architecture for AI.
Fourth, I’ll explore how the entire industry chain and R&D team structures need to evolve to adapt to these changes. Finally, I’ll share my outlook on EE architecture. Although it is currently applied in the automotive sector, it can eventually empower the entire ecosystem of embodied intelligent products, including humanoid robots, drones, flying cars, and other smart embodied devices.
Q2
汽车 EE 架构一路走来经历了哪些演进阶段？AI 浪潮来袭，会给 EE 架构带来什么样的冲击？后续架构演进又会朝着什么方向发展？
What are the evolutionary stages of automotive EE architecture? How will the AI wave impact it, and in what direction will the architecture evolve in the future?
蒋峰 :
我从事EE架构这个行业已经差不多快10年了。十年前，大概是在2016年的时候，就见过博世提出的一个EE架构路线图，相信很多同行也在不同场合引用过“从分布式到中央架构的演进路线”。最早实现中央区域架构的是特斯拉的Model 3，大概在2016或17年左右就已经提出了中央加区域的架构，而且是在Model 3车型上应用了，在当时是很先进的架构。到现在来看，国内的车企现在中央加区域这种架构已经进入主流时代，特别是新能源汽车，现在基本上都采用了这种架构形态。中央加区域架构的优势就是打破了传统的功能域界限，以物理区域实现集成，有利于算力的整合，而且兼顾了线束的最优布局。
现在到了AI时代，和我们之前的架构本质上的变化就是出发点变了。以前架构开发都是以功能需求为基础，基于V模型的方式，从需求到分解再到详细设计再到开发测试，是一种基于规则的方式去开发。AI时代来了之后，我们很大的特征就是依赖于数据和模型，这样对于整个架构有很大变化。我们需要大量数据来支撑AI，大量算法来支撑模型实现。这就对架构层面提出要求：大数据需要通讯的高速传输、低时延，还有高算力的处理器。现在的AI模型相当复杂，而且到了多域融合阶段，还要处理不同功能安全等级、不同功能域的算法，这对高算力处理器的资源调度也有更高要求。
我们怎么应对呢？重点一个是通讯技术，已经从传统的CAN通讯几兆，到现在的以太网逐步成为主干网，已经达到了1G甚至10G的带宽。高算力的SoC也不断进入视野，芯片技术和通信技术的发展为我们后面的AI架构发展带来了契机。后续架构演进会继续朝着中央大脑加传感执行的形态发展，最终趋同于人的架构，大脑指挥四肢。
Mr. Jiang Feng :
I have been working in the EE architecture industry for almost 10 years now. Ten years ago, around 2016, Bosch presented an EE architecture road map that many peers have cited ‘the evolution from distributed to central architecture’. The first to implement the central+zone architecture was Tesla Model 3, around 2016 or 2017, which was very advanced at that time. Now, domestic car manufacturers have made central+zone architecture mainstream, especially for new energy vehicles, which basically all adopt this form. The advantage is that it breaks traditional functional domain boundaries, achieves integration based on physical zones, enables better computing power consolidation, and optimizes wiring harness layout.
Now we have entered the AI era. The essential change is that our starting point has shifted. Previously, we developed based on functional requirements using a Vmodel approach – a rulebased rigorous process. In the AI era, we rely on data and models. This brings significant changes to the architecture: we need massive data to support AI, and a large number of algorithms to implement our models. At the architecture level, big data requires highspeed, lowlatency communication and highcomputingpower processors. Current AI models are quite complex, and at the multidomain integration stage, we also need to handle different functional safety levels and algorithms from different domains, imposing higher requirements on resource scheduling of processors.
How do we respond? Our focus is on communication technology – from traditional CAN at a few megabits to Ethernet gradually becoming the backbone network, now reaching 1G or even 10G bandwidth. Highcomputing SoCs are also emerging. The development of chip and communication technologies brings opportunities for our future AI architecture. The architecture will continue to evolve toward a central brain plus sensors and actuators, eventually converging toward the human architecture – the brain commands the limbs.
Q3
站在工程实践角度，AI 上车之后，整车功能开发和以往相比，在工作模式、开发流程上出现了哪些改变？
From an engineering practice perspective, what changes have occurred in vehicle function development working models and processes after AI is onboard?
蒋峰 :
现在的开发整个流程是基于AI的，我们可以实现很敏捷的开发模式。这个架构会把传感和执行部分，类似于人的四肢，逐步标准化。那么后面的开发主要就是开发大脑的算法，这是开发侧重点的变化。另外，在整个流程体系方面，我们也借助AI来提升效率。
以前开发软件，我们都是人通过编程语言去写，进入AI时代后，可以通过AI工具来帮我们写代码，极大提升了效率。到集成和测试阶段，我们也会借助自动化AI工具来提升软件集成和功能测试的效率。这些都是AI时代下的变革。
另外，整个AI时代因为发生了这些变革，实际上对我们传统的开发工程师的能力有了更高的要求。传统分工比较细，做硬件、做软件、做开发测试的。现在要做好一个AI架构，确实需要复合型人才，才能把整个AI架构到开发做到最极致的结果。
Mr. Jiang Feng :
Our current development process is based on AI, allowing us to achieve a very agile development model. This architecture will gradually standardize the sensing and actuation parts, similar to our human limbs, so our subsequent development will mainly focus on developing the algorithms for the brain. That’s a shift in development emphasis. In addition, we leverage AI to improve efficiency across the process system.
Previously, we humans wrote code using programming languages. Now in the AI era, we can let AI tools write code for us, greatly improving efficiency. During integration and testing, we also use automated AI tools to improve software integration and functional testing efficiency. These are the transformations in the AI era.
Furthermore, these changes actually place higher demands on the capabilities of our traditional development engineers. Traditionally, the division of labour was finegrained- hardware, software, development, testing were separate. Now, to build a good AI architecture, we really need versatile talents so that we can drive the entire AI architecture development to the best possible result.
Q4
智能驾驶和智能座舱融合已经成为行业热点，这种融合能够带来哪些优势？当前又有哪些技术难题阻碍融合落地？
The integration of intelligent driving and smart cockpit has become a hot topic. What advantages does this integration bring, and what technical challenges are currently hindering its implementation?
蒋峰 :
今年智能座舱和智能驾驶进行融合是非常热门的话题，头部芯片供应商也逐步推出了一些舱驾一体的高算力芯片。但整个舱驾融合的发展现在还处于博弈阶段。舱驾融合最大的好处就是算力资源共享。以前是两颗不同的高算力SoC分别负责舱和驾，算力冗余分散，共用一颗芯片后实现资源共享，对硬件资源节约显著。
另外，舱和驾之间有很多交互功能，以前通过外部总线如以太网或SERDES传输，舱驾融合后数据共享在芯片内部，能提升数据调用的效率。从这两方面看，一是资源节约降低成本，二是软件融合提升开发效率。
但舱驾融合还存在瓶颈。首先，舱和驾有各自特征：自驾逐步向高阶等级发展实现不同功能体验，座舱更多倾向于生态应用方向。
一个是强调高安全价值，一个是高体验生态，发展是两条不同路线，各有节奏。融合在一起就面临搭配组合的问题，选择舱驾一体芯片时要思考什么梯度的智驾芯片搭配什么梯度的座舱芯片，搭配合适才是用户可接受的最好产品，搭配不合适会浪费开发资源。
第二个问题是舱驾融合不仅需要硬件芯片支撑，同样需要开发团队的融合。国内很多主机厂像长安一样，自驾和座舱开发团队是两个独立部门，要进行很好的舱驾融合，整个开发团队和流程体系都需要整合重构。通俗讲就是要打破部门墙进行变革，有一定难度，需要时间才能做得更好。
Mr. Jiang Feng :
This year, the integration of smart cockpit and intelligent driving is a very hot topic. Leading chip suppliers have launched highcomputingpower chips for cockpitdriving integration. However, the development is still in a stage of contention. The biggest benefit is the sharing of computing resources. Previously, we had two separate highcomputing SoCs for cockpit and driving, with scattered redundancy. After sharing one chip, we achieve resource sharing and significant hardware cost savings.
Also, there are many interactive functions between cockpit and driving. Previously, we used external buses like Ethernet or SERDES to transmit; after integration, data sharing happens inside the chip, improving data access efficiency. So, from resource savings (cost reduction) and software integration (development efficiency improvement) , these are the benefits.
But there are bottlenecks. First, cockpit and driving each have their own characteristics. Autonomous driving gradually moves toward higher levels to deliver different functional experiences; cockpit leans more toward ecosystemoriented applications.
One emphasizes high safety value, the other highexperience ecosystems. They are two different development paths with their own paces. When merging them, we face the matching problem: when choosing a cockpitdriving integrated chip, we need to think about what level of ADAS chip to pair with what level of cockpit chip. Only with a proper match can we deliver the best product users will accept. An improper match wastes development resources.
Second, integration requires not only hardware chip support but also integration of development teams. Many domestic OEMs, like Changan, have autonomous driving and cockpit teams as two separate departments. To achieve good integration, the entire development team and process systems need to be integrated and restructured. To put it plainly, we need to break down departmental silos and drive change, which is challenging and takes time.
Q5
汽车 EE 架构和人形机器人这类具身智能产品的 EE 架构有哪些相似点？如果实现 EE 架构跨品类复用，能产生哪些价值？
What are the similarities between automotive EE architecture and the EE architecture of embodied intelligence products such as humanoid robots? If EE architecture can be reused across product categories, what value can it bring?
蒋峰 :
现在汽车已经发展到高度智能的时代，它不只是一个简单的交通工具，而是具有智能出行的工具，可以作为你的助手帮你完成很多事情。我们整个汽车可以称为具身智能的产品。类似的还有飞行汽车、人形机器人等，它们与智能汽车有一些相似之处，比如对外界事物的感知通过摄像头和雷达，经过中央大脑处理，最后到执行。只是执行机制有差异：汽车执行主要是行驶、动力、制动、转向；机器人执行更多是类人的手脚关节等。相似之处在于架构上都有大脑和执行机构，有很大机会去复用。
复用汽车EE架构到其他具身智能产品，最大的好处是得益于汽车产业链的成熟度。主流车企年产量达到上百万辆规模，相对于人形机器人、飞行汽车这些刚起步的产业，规模量是几十倍甚至上百倍的关系。如果直接复用汽车里面的芯片形态，会大幅降低这些产品的硬件成本。第二个好处是汽车架构里的OTA升级、安全策略、电源管理等技术也可以复用到人形机器人等产品上，开发效率会极大提升。
虽然架构可以赋能其他具身智能产品，但人形机器人、飞行汽车还是有各自特征，比如人形机器人的控制精度，飞行汽车的安全性等方面有差异。只有做到这些方面的坚固性考虑，才能把架构复用做到最好。
Mr. Jiang Feng :
Today, automobiles have evolved into a highly intelligent era. They are no longer just simple means of transportation, but smart mobility tools that can assist you in many tasks. Our entire vehicle can be considered an embodied intelligence product. Similar products include flying cars and humanoid robots, which share some similarities with smart cars. For instance, they perceive the external environment through cameras and radars, process the data via a central brain, and then execute actions. The main difference lies in the mechanisms involved. In cars, these mechanisms primarily relate to driving, propulsion, braking, and steering. In robots, they are more focused on human-like joints in the arms and legs. The similarity is that both have a brain and actuation systems, so there is a great opportunity for reuse.
Reusing automotive EE architecture in other embodied intelligence products offers several major benefits. First, thanks to the maturity of the automotive industry chain, mainstream automakers produce millions of vehicles annually. Compared with emerging industries like humanoid robots and flying cars, which are still in their infancy, the production scale is tens or even hundreds of times larger. By directly reusing the chip form factors from automobiles, hardware costs for these products can be significantly reduced. Second, technologies such as OTA upgrades, safety strategies, and power management in automotive architecture can also be reused in humanoid robots and other products, greatly improving development efficiency.
Although the architecture can empower other embodied intelligence products, humanoid robots and flying cars still have their own characteristics. For example, humanoid robots require higher control precision, while flying cars have more stringent safety requirements. Only by addressing these specific robustness considerations can architecture reuse be optimally achieved.
关注我们