After a decade of building cars, new energy vehicle makers have generally moved from lightweight engineering models to more technically integrated vehicle platforms. Li Auto is trying to make that shift explicit.
The company has declared an ambition to become an “embodied intelligence company” while building more of its whole-vehicle technology stack in-house. Over the past four years, it has rebuilt much of its chassis system, working with suppliers to co-develop three major technologies: electromechanical brake-by-wire, 800-volt active suspension, and steer-by-wire.
Li Auto is also expanding in-house battery production. In range extension technology, the system that underpinned its early success, the company has moved further into internal R&D and production. Its latest-generation range extender, effectively the engine, is now produced by Li Auto itself.
Those capabilities will be concentrated in its latest flagship vehicle, the Li L9 Livis.
For Li Auto, the new Li L9 is also intended to mark the starting point for its next generation of vehicle technology.
“From day one of project approval, we set a goal: the suspension and chassis of the new Li L9 had to be the best in the industry,” said CEO Li Xiang when the Li L9 Livis made its first appearance at the Beijing Auto Show. During a launch presentation that lasted only 20 minutes, he spent half the time discussing the chassis and suspension systems.
In an interview with Liu Liguo, head of vehicle electric R&D at Li Auto, 36Kr was told that Li Auto built its chassis strategy around three considerations: creating a generational lead in user experience, developing a coordinated chassis execution layer as intelligent vehicle systems evolve, and strengthening supply chain security.
Liu added that it is a “major misconception” to think Li Auto does not value technology, saying the opposite is true.
Automakers have spent years improving cabins, assisted driving systems, and batteries. But perception and decision-making still need to be translated into physical movement. That is where the chassis becomes central.
“Without coordinated chassis execution, even the smartest driving assistance system cannot stay steady,” Liu said. The chassis domain, which coordinates vehicle motion, is the execution layer that Li Auto believes is needed to turn intelligent driving decisions into controlled movement.
Li Auto believes that fully connecting chassis capabilities requires control over active suspension, brake-by-wire, and steer-by-wire. “Only by putting these systems together can we create a gap in both technology and experience,” Liu said.
In the second half of 2021, Li Auto began R&D on 800V active suspension and a full drive-by-wire chassis. The most technically demanding of these efforts, according to Liu, was electromechanical braking. At the time, the industry chain was immature. Liu said the technology existed mainly at the proof-of-concept stage, with no mass production precedent and no clear regulatory framework.
As a next-generation chassis technology, electromechanical braking, or EMB, eliminates the hydraulic braking system long used in the automotive industry. It removes brake fluid, hydraulic pumps, and complex lines, instead using independent motors at each of the four wheel ends to drive caliper braking directly.
After four years of development, Li Auto has implemented three major technologies: 800V active suspension, steer-by-wire, and brake-by-wire.
At the hardware level, Li Auto worked with suppliers on joint R&D. At the software level, Li Auto engineers developed the systems internally.
“Many OEMs (original equipment manufacturers) have also procured steer-by-wire and brake-by-wire, but the software for the three systems comes from black boxes provided by different suppliers. The code cannot communicate. There is no way to achieve unified control,” Liu said.
To provide enough computing resources for chassis algorithms, Li Auto designed a dedicated vehicle control area inside the Mach M100 chip and reserved 100–200 TOPS (tera operations per second) of compute. The company said this gives the chassis execution system enough compute capacity to keep pace with assisted driving decision systems.
Li Auto’s track record has often been attributed externally to product definition, especially its family-oriented positioning and the refrigerator, television, and sofa features that became shorthand for its interiors. But as rivals have copied those visible features and the company’s sales have come under pressure, Li Auto needs to show that it has a deeper technical moat.
The following transcript has been edited and consolidated for brevity and clarity.
36Kr: Li Auto’s biggest chassis breakthrough this time is in EMB. When did you sense this coming?
Liu Liguo (LL): Around the eve of the L series launch in 2022, that generation of technology had entered mass production, and we began reserving technologies for the next generation. We predicted that the next breakthrough direction would be comprehensive electrification and smart enhancement of the chassis.
Only after the chassis becomes electrically controlled can there be more room for intelligence. Past electromechanical systems were extremely complex, rule-based control was very cumbersome, and each company developed systems independently, making integration into a single whole extremely difficult. We believed that chassis electrification was inevitably the future direction, so we gradually began reserving technologies in three areas: steering, braking, and suspension.
36Kr: Did Li Xiang set any requirements for this generation of chassis R&D?
LL: Li Xiang’s biggest requirement for the next-generation flagship model was that it had to deliver a completely different driving and riding experience and create a technology generation gap. That was his biggest requirement. From requirements definition to human interaction and final acceptance, he participates in every stage. He often test-drives the car, basically once every two or three weeks.
36Kr: Other OEMs had already launched active suspension and steer-by-wire. Did Li Auto consider putting some technologies onto facelift models first?
LL: To assess the driving and riding experience, we had to evaluate the entire chassis domain together, including the 800V active suspension and full drive-by-wire chassis, rather than looking at a single point. Looking at any one system alone has limited value. For example, steer-by-wire alone basically enables variable steering ratio, but that alone is not significant enough. Once combined with active suspension and rear-wheel steering, its value is amplified and steering also becomes easier.
When the three systems coordinate, the value is an overall feeling where the addition of three ones exceeds three. Only by putting these systems together can we create a gap in both technology and experience. That is what makes it a full drive-by-wire chassis with real potential for embodied intelligence.
36Kr: Before working on EMB, did Li Auto explore other routes?
LL: Around 2022, the mainstream view was that the mass production window for EMB would be 2028–2030, or even later. Regulations were completely blank at the time. Brake-by-wire regulations only began taking effect on January 1 this year. Nobody could tell you whether this route would ultimately work.
At the time, there was one approach that had a strong voice: the “wet front, dry rear” hybrid braking route. The front two wheels retained hydraulic braking, while the rear two wheels used a dry solution. This route was relatively conservative. If something went wrong with the rear wheels, the front hydraulic system could still provide a fallback. This was once believed to be the mainstream transitional direction.
36Kr: Why did Li Auto choose EMB over this approach?
LL: We discussed this internally, but after our engineers reasoned from first principles, they found several fatal problems. The wet front, dry rear route could not achieve EMB’s real advantages. The software platform could not move up from the wheel end, software and hardware still could not be decoupled, and the electrohydraulic hybrid control made the whole system harder to maintain.
36Kr: Drive-by-wire chassis has been an area of emphasis in recent years, but user perception is not very explicit. What do you think is EMB’s real value?
LL: EMB’s real value lies in independent four-wheel control and the complete upward migration of software. Only execution remains at the wheel end, while decision-making and coordination algorithms all move upward. The wet front, dry rear route cannot achieve either. It is equivalent to paying the development cost of EMB while getting less than half the benefits.
In terms of safety, EMB is safer than the previous electrohydraulic braking system. With the same tires and the same road adhesion, EMB’s braking distance is more than two meters shorter than EHB. The anti-lock braking system (ABS) can be tuned more precisely, making better use of adhesion. At the same time, the autonomous driving control loop is very short, so driving feels better.
So we ultimately went directly to fully dry EMB. After we built the initial sample, other manufacturers sensed the direction and began following up. Notably, after many leading global suppliers learned that our EMB had succeeded, they also approached us proactively to seek cooperation.
36Kr: Are there testing metrics? What evaluation standards are there for failure?
LL: There are clear regulations for EMB. Fully loaded deceleration is assumed to be one. When it falls below 0.65, there must be an alarm and a corresponding warning mechanism. The system must know what has failed and have a corresponding solution. Any loss-of-control scenario cannot fall below 0.25. The regulations are clear on all of this.
Since implementation, we have not had a single case of deceleration falling below 0.65.
36Kr: What does 0.65 mean, in deceleration terms?
LL: In ordinary driving, people are usually at 0.2 or 0.3. For many people, emergency braking during driving is around 0.6. Unless you have been trained or press the pedal all the way down, you may reach around one.
36Kr: The main supplier for EMB this time is Bethel. Did Li Auto consider working with international suppliers?
LL: International suppliers such as Bosch and ZF have worked deeply in electro-hydraulic braking (EHB) for decades. Relying on global platform scale, their technology is mature, costs are extremely low, and they already have stable commercial returns in the existing track. For the completely new EMB route, their investment pace did not fully match our R&D needs. This is not a question of capability, but a difference in commercial drive.
Large suppliers have complex processes. If you approach them, they will develop with you, but the urgency, pace, and level of investment will not necessarily follow your rhythm. That is also why we wanted independent R&D. China’s decision to issue EMB regulations ahead of the rest of the world also reflects a desire to achieve self-controlled component supply and leapfrog through a new route in this field. So we later chose Bethel to develop EMB together.
36Kr: Discussion has been rife about the 48V and 800V active suspension routes. How does Li Auto see this? Why choose 800V active suspension?
LL: 48V was first promoted in Europe and the US to serve traditional fuel vehicles. It is relatively simple, but it cannot use high voltage.
In the early development process, suppliers initially also wanted to recommend 48V. But we found that the power 48V can provide is relatively small. To achieve high power, current must be increased, and the wiring harness has to be very thick to carry it.
From the demand side, what is the core goal of active suspension? It is to allow the four wheels to truly respond independently to the road surface and driving dynamics, without compromising with one another. The prerequisite is that you must have enough active force. You need to be able to actively apply enough force to each wheel to control it. Only then can you talk about decoupling.
A 48V system has an upper limit in power, and the active force it can provide is around 7,000 newtons. For an SUV, that is not enough. During high-speed cornering, roll moment is large. If 6,000 N is not enough, you have to retain the anti-roll bar for assistance. Once the anti-roll bar remains, the left and right wheels are connected. When one wheel moves, the other is also affected. You cannot achieve independent four-wheel control.
We also studied several routes on the market at the time, including linear motors, ball screws, and 48V hydraulic pumps. We disassembled and analyzed all of them. Each had its logic, but each also had a ceiling. We ultimately chose the 800V hydraulic pump not because it was the “most advanced,” but because it was the only solution that could truly achieve our goal. The 800V solution can push active force above 10,000 N, which is necessary to meet our goals for multiple user scenarios, including single-wheel lift.
36Kr: What validation did Li Auto conduct before putting this new technology into a vehicle for the first time?
LL: Li Auto’s dedicated drive-by-wire chassis road tests have accumulated more than four million kilometers, covering various climates and operating conditions across 23 provinces and cities nationwide. In all kinds of extreme scenarios and simulated tests involving manually disconnected signals, there has been no total failure.
The EMB system’s functional safety has received ASIL D certification from DAkkS, the highest international standard. Measured braking distance reached 33 meters, comprehensively surpassing traditional hydraulic braking systems. National regulations are the baseline. Our standard is far above that baseline.
36Kr: Li Auto calls the L9 Livis an embodied intelligence flagship SUV. What part of embodiment does the chassis domain handle?
LL: The whole coordination and kinematic control layer is the vehicle’s motion execution system. Assisted driving is the decision system.
The core requirement that embodied intelligence places on the chassis is that every joint must truly follow commands independently. In other words, all four wheels must be fully decoupled across the X, Y, and Z axes.
Suspension achieves full four-wheel decoupling on the Z-axis. Each suspension independently controls support force, which enables scenarios such as single-wheel lift. Steer-by-wire and rear-wheel steering achieve Y-axis decoupling. The steering wheel and wheels rely entirely on electrical signals, and the physical connection is broken. EMB is the final piece. Each of the four wheels has its own electromechanical actuator, with no hydraulic or mechanical connection. It is physically independent in the true sense.
Only when decoupling is achieved in all three directions, and when the entire chassis domain achieves deep coordination, can embodied intelligence ultimately be realized.
36Kr: How do these three systems coordinate deeply?
LL: Behind them is an even deeper wall: software. Many OEMs have also procured steer-by-wire and brake-by-wire, but the software for the three systems comes from black boxes provided by different suppliers. The code cannot communicate. There is no way to achieve unified control.
The software for steering, braking, and suspension is all developed in-house by us. The code sits within the same architectural system, and data is naturally connected. Only in this way can the domain controller truly coordinate the three directions, instead of passing messages through black boxes.
36Kr: What changes does this bring to actual experience?
LL: Based on a high-precision nonlinear full-vehicle model, the system predicts vehicle motion trends in real time and issues coordinated commands to the three systems in a unified way. It does not wait for the body to lose balance before correcting it. It intervenes in advance, so steering, braking, and suspension all apply force toward the same target at every dynamic moment.
During emergency avoidance, stability control intervenes 300 milliseconds earlier than traditional solutions. End-to-end latency is compressed from more than 100 milliseconds to within 30 milliseconds. It is equivalent to an execution system in the human body. You do not perceive it working, but it is always there.
This is not only about making the car drive better. Level 4 autonomous driving and embodied intelligence are essentially a complete closed loop of perception, decision-making, and execution. The driving assistance system is responsible for perception and decision-making, but instructions ultimately have to become real body motion. If the chassis does not have a coordinated control center that can receive commands uniformly and respond within milliseconds, assisted-driving instructions have no executor.
Without coordinated chassis execution, even the smartest driving assistance system cannot stay steady.
36Kr: How does the chassis execution system coordinate with decision-making of the driving assistance system?
LL: Coordination between both systems is already closely integrated in downstream trajectory planning.
In previous control algorithms, the action in vision-language-action (VLA) output a trajectory, the trajectory was converted into acceleration, the chassis accepted it, and then it was converted into force. Now it has become one model. The trajectory can go directly to force, reducing one more link compared with before.
36Kr: The earnings call previously mentioned that the Mach chip integrated the XCU domain controller. How was that done?
LL: Inside the chip, there are roughly dozens of cores. There is a dedicated area for assisted driving and another area for vehicle control, among others.
Vehicle control was involved from the first day of chip design. Vehicle control, assisted driving, and other systems have different iteration cycles, safety levels, and latency requirements. It is difficult to handle them through time-sharing control or virtual machines, so physical hard isolation was designed at the chip design stage. It is like small rooms, each separated by partitions.
36Kr: How much chip compute was allocated to the chassis?
LL: Around 100–200 TOPS. In a traditional OEM, that 100–200 TOPS would already be comparable to the entire AD compute capacity of some other automakers.
Currently, only Tesla has done this kind of integration more extensively. Li Auto is the second after Tesla to reach this level.
36Kr: There was an impression that Li Auto did not emphasize technology. Is that a misconception?
LL: I think it is a major misconception. In fact, many things Li Auto does are supported by very deep technical and software capabilities. For example, we were one of the earliest companies in the market to lay out 5C charging. We self-developed range extension systems and electric drive systems. From chips to packaging and assemblies, we insist on in-house R&D and design. Behind these capabilities are very deep technical accumulation and vertical integration capability across the entire industry chain.
Looking across China, how many companies can really do this?
It is just that, in external communication, Li Xiang prefers to use product language that users can understand, converting complex technical capabilities into expressions that are easier for users to understand. Precisely because of this, outsiders sometimes mistakenly think Li Auto does not value technology. But the opposite is true.
36Kr: Will today’s chassis capabilities migrate to robots?
LL: Automotive chassis and robotics technologies share the same origin, especially at the control level. From the underlying OS to algorithms, from chips to vehicle applications, everything is developed in-house. This technology system provides a solid foundation whether we are building cars or robots. From the OS to the base layer and chip deployment, it is all our own. That is the premise behind automakers entering robotics.
36Kr: After the chassis domain becomes tightly integrated with assisted driving, will its software layer also follow the scaling law?
LL: I do not think the accumulation of chassis data contributes much to chassis algorithms.
Data accumulation in assisted driving follows the logic of scaling law: more data, more compute, larger models, better functions. But chassis control is model-based control. This model is not a large language model. It is a vehicle control model based on vehicle dynamics, a mathematical model of the physical world, and it does not require excessive data.
For chassis, the point that really creates differentiation is the dynamics model. Whether the full-vehicle model is more precise and more intelligent depends fundamentally on the capability of this mathematical model.
36Kr: In integration, is battery-chassis integration, such as cell-to-body (CTB) or cell-to-chassis (CTC), something Li Auto wants to do next?
LL: If we make sedans in the future, we will definitely consider it, because it contributes to interior space along the Z-axis, meaning height. CTB’s greatest value lies there. In sedans, second-row headroom is very tight, around 920–930 millimeters. After integration, an extra ten millimeters makes a very large contribution.
36Kr: What will the next R&D direction for the chassis domain be?
LL: There will be small hardware iterations, but there will not be significant changes in form. Whether active suspension, steer-by-wire, or EMB, local parts will continue to iterate to make reliability higher. The upper limit for reliability has no endpoint. The pursuit is always zero PPM, or zero defects per million.
The core remains the OEM’s system integration capability: integrating the execution systems in the three directions through a full-vehicle model to coordinate control of the three forces. The emphasis is more on control, making the whole driving and riding experience better. Another direction is integration with AD, making the assisted driving experience better, response faster, and safety performance stronger.
Personalization is also an important direction for our next generation of R&D. The system will adapt to your driving preferences, or the driving preferences you want. You can tell it what you like, and it will report its current state. If you say this is not what you want, and specify the dimensions where you want it to be gentler or more aggressive, you can adjust it, and the car will adapt to you.
36Kr: How does Li Auto internally judge the value of technology and products?
LL: Internally, we have a product world model. It is not the world model used in artificial intelligence simulation. It is similar to the integrated product development (IPD) process. In one vehicle, it contains the core selling points, key technologies, business aspects, and so on.
36Kr: How does Li Auto balance innovation investment against wastage brought by uncertainty?
LL: If you are doing forward-looking research, you will inevitably face detours. From the chassis perspective, however, among the technologies we selected, Europeans have worked on many of them over the past 100 years. So in this regard, there are not that many detours.
36Kr: The all-new Li L9 Livis will be delivered soon. What expectations does Li Auto have for the car?
LL: It is the world’s first vehicle to mass produce electromechanical brake-by-wire, steer-by-wire, rear-wheel steering, and 800V active suspension together. It is also the first vehicle to allow these four systems to truly coordinate through full-stack in-house algorithms. Having the hardware in place is only the prerequisite. Connecting the software is the core.
Once the chassis truly achieves comprehensive electronic control, many things that could not be done in the past have only just become possible: deep coordination between assisted driving and the chassis, personalized driving adaptation, and more extreme safety redundancy. All of these need to continue iterating on the basis of full drive-by-wire.
Internally, we say that the past four years were about teaching the chassis to think with electricity. Next, it is about teaching the chassis to help people think. The day the Li L9 Livis reaches users is not the endpoint. It is the first day this truly begins.
KrASIA features translated and adapted content that was originally published by 36Kr. This article was written by Xiao Man for 36Kr.