The ambition of the "whistleblower" of 全民智驾 (Universal Smart Driving)

Author | Chai Xuchen

Editor | Wang Xiaojun

As the intelligent vehicle competition initiated by new forces reaches a fever pitch, Changan Automobile has taken the lead as the whistleblower for universal intelligent driving, completely overturning the market's perception of the "national team."

Twenty days ago, Changan unveiled its ambitious "Beidou Tian Shu 2.0" intelligent strategy in its home ground of Chongqing, with the key goal of achieving equality in intelligent driving, equipping national-level models priced around 100,000 yuan with lidar.

Changan Automobile's Chief Intelligent Driving Technology Officer, Tao Ji, told Wall Street Insights that intelligent driving has indeed reached a turning point, with its capabilities and performance gradually reassuring users. As costs decrease and public awareness increases, intelligent driving will rapidly become widespread.

However, intelligent driving is just the tip of the iceberg of Changan's ambitions. According to internal planning, it aims to reach L3 and L4 levels of intelligent driving by 2026 and 2028, respectively, with plans for manned flying vehicles and embodied robots also in the works.

Tao Ji stated that Changan is undergoing its third entrepreneurial transformation, evolving from fuel vehicles to electric vehicles, and now to becoming a smart low-carbon mobility technology company. This aligns with the visions of emerging players like XPeng, Li Auto, and Tesla.

It is evident that Changan is striving to shed the inherent characteristics of traditional automakers, using AI to connect its grand ambitions and "aggressively" responding to new forces.

However, the transforming Changan is not going it alone; besides Huawei, Tencent Cloud serves as a "key helper" quietly providing the underlying infrastructure for its AI strategy.

In discussions with Tao Ji, he frequently mentioned data, stating, "This is the most important nutrient in the era of AI. How to fully acquire data? How to manipulate it is crucial."

Indeed, as autonomous driving capabilities upgrade, a massive "data tsunami" has emerged. Industry competition has extended from vehicles to the cloud, making high-performance, low-cost, safe, and compliant data storage, computation, and model training in the cloud a necessity, further expanding the automotive industry's cloud market size.

In response to the needs of automotive companies for autonomous driving research and operations, Tencent Cloud has taken the lead in establishing dedicated "smart automotive cloud" zones, creating an end-to-end, fully compliant data closed-loop service for autonomous driving development. Tencent has also developed the Intelligent Driving Cloud Map, fully cloudifying autonomous driving map data to provide map data "nutrients" to autonomous driving systems in a more flexible and dynamic manner.

Tang Daosheng, Senior Executive Vice President of Tencent Group and CEO of the Cloud and Smart Industry Group, has pointed out that Tencent Cloud's efforts aim to allow Changan and other automakers' R&D teams to focus more on algorithm development.

Of course, with the explosive demand for intelligent driving and intelligent cockpit solutions built on AI large models, there has been a surge in demand for cloud computing power. The future competition in intelligence will fundamentally be a competition for computing power. XPeng Chairman He Xiaopeng even predicts that starting in 2025, XPeng will spend over 1 billion yuan annually on computing power This means that as automakers compete for the high ground of intelligence, Tencent Cloud's layout in the automotive field has finally ushered in the moment of AI scenario implementation. This secures a greater winning edge for it in the fierce battle for public cloud market share.

Currently, Tencent Cloud is showcasing its imagination to become an AI infrastructure provider in the automotive industry, and its partnership with Changan may just be a model room. How will the two parties collaborate to stir up the automotive circle and even the artificial intelligence track?

The following is a transcript of the dialogue with Changan Automobile's Chief Intelligent Driving Technology Officer Tao Ji, Deputy General Manager of Changan Automobile's Intelligent Research Institute Mao Yeping, and Deputy General Manager of Wutong Technology Liu Tongyang (edited):

Q: What new changes and achievements has Changan made in intelligence over the years?

Mao Yeping: We started in 2015, and by 2018 we launched the country's first automatic parking APA 5.0. Last year, we established a central ring network architecture that only Tesla has, and under the dual protection of this architecture, we have evolved into smart vehicles.

From the seven to eight years of accumulation, we have gradually transformed the car into a six-layer architecture. After proposing the concept of smart vehicles, it is more suitable for the evolution of autonomous driving now, because many perception and decision-making actions are coordinated with the rapid response of all related components of the car, while the response speed of traditional methods is relatively much slower.

In this process, we have many software partners, such as Tencent, Horizon, Huawei, and ZTE. As you know, the previous intelligent driving technology routes included Tesla's FSD and high-precision maps, with many directions. In the face of uncertain directions, only partners can come together.

Changan should be considered a technology integration faction. In the process of integration, all parties need to bring their best technologies together.

Since February 2023, we have been committed to equalizing intelligent driving. We have already built the basic version of Changan's intelligent driving on all our models by the third quarter of last year. We were the first to lead the way in intelligent driving and have now achieved equalization, with features like remote parking and one-click parking already popularized.

In the past, the question of whose intelligence was better or worse relied heavily on the strength of their suppliers, such as Bosch, Continental, ZF, or related hardware providers, while the core of the vehicle enterprises was in integration. At that time, it was said that the core strength of automotive manufacturers lay in their integration capabilities.

Since 2021, we announced the SDA architecture at the first Technology Changan Conference, which we built from scratch and is absolutely leading in the industry. We also divided the car into six layers.

The benefit of this is that in an era where intelligent updates and iterations are so frequent or rapid, if we still rely on suppliers to provide these capabilities, which are all black boxes, how can we keep up with user demands? In the future, everyone will see what central enterprise intelligent driving is, and the momentum of this batch of central enterprises in intelligent driving may bring new shocks to everyone.

Q: How does Changan balance self-research and cooperation in intelligent driving capabilities?

Mao Yeping: We walk on two legs; self-research focuses on deep integration, while partners expand the ecosystem after being integrated into the vehicle. In the industry, the difficulties still lie in perception, decision-making, and breakthroughs in computing power; the technical common problems are the same We invested 11.5 billion in Huawei to jointly develop advanced technologies. There are solutions for high-end users. For ordinary people, how to make everyone feel the convenience that technology brings is crucial. For cars priced over 200,000, actually 80% of users have not enjoyed the convenience brought by intelligent driving or advanced intelligent driving; consumer habits have not been cultivated. Therefore, we are also taking another approach, with Changan TianShu focusing on equalizing access to advanced intelligent driving.

We also hope to work together with our competitors and Chinese brands to cultivate user habits collectively. Once you experience it, it's hard to go back to the past.

Q: If universal intelligent driving is realized, how will the relationship between automakers and suppliers evolve?

Tao Ji: First of all, I believe that in the long term, universal intelligent driving essentially still generates value through intelligent driving, and this value must create a willingness to consume among users for it to be sustainable in the long run. The product strength of intelligent driving itself must be established, making users feel that it is useful, enjoyable, and helpful in their daily lives, generating a willingness to pay. When your payment exceeds its cost, it can operate.

In the short term, there are many things to do—cost reduction. From the perspective of the supply chain, the replacement of domestic chips is already in use, which is a significant cost reduction. How to maintain good performance with the original algorithms while replacing with domestic alternatives is a technical challenge.

The domestic lidar supply chain is also rising; currently, mainstream vehicle-mounted lidar has dropped below 1,000, which is also a result of technological changes, as large production volumes can reduce costs.

We are currently working on some deep customization of lidar, migrating calculations to the central domain controller. Additionally, retaining more and richer received signals, information retention and processing is a dual-purpose task, but this requires companies to deeply understand and innovate in hardware and have the capability to participate in this area.

Q: What conditions need to mature for universal intelligent driving to become a collective slogan or action among automakers this year?

Tao Ji: I think it still starts with the maturity of intelligent driving technology itself; eventually, everyone will love using intelligent driving. A few years ago, the cruise and navigation functions on highways could only be said to exist, but users were not very confident in using them, and some were even discouraged from using intelligent driving because once they tried it, they were afraid to use it again.

However, now people clearly feel that if they have experienced the performance of top-tier advanced intelligent driving on highways, it is actually quite perfect, including following cars with acceleration and deceleration, efficient overtaking, and the ability to navigate on and off ramps, which makes people feel very secure, even for hundreds of kilometers of travel without any human intervention.

When you have this capability, everyone will feel that it is indeed a helper, reducing fatigue and making driving easier. Once its costs come down and people's understanding increases, it will become a popular product. Therefore, automakers will believe that this year, advanced or mid-to-high-level intelligent driving can achieve universal adoption.

In this significant technological phase, I believe it may be a good opportunity to bridge the gap and allow more users to understand and use intelligent driving products. We are "betting" that this time point will arrive this year Question: With end-to-end technology being widely emphasized now, will there be a change in the timeline for Changan's intelligent driving large-scale deployment?

Tao Ji: The initial consideration last year was to approach it in two steps: first focusing on highways and then urban areas, completing the highway aspect within a year and tackling urban areas by 2025. We are still progressing according to this plan.

However, during this process, we have seen new technologies continuously emerging, including end-to-end and large model technologies that need to be integrated into vehicles. Changan's response has been very quick; we started discussing the establishment of end-to-end projects and talent reserves in the first half of last year, and we completed the project establishment in Q3 of last year.

The overall end-to-end project team is dedicated to rapidly advancing research and development while also adjusting the organizational structure. We are integrating and merging traditional perception and control with the new end-to-end paradigm, focusing on how to adapt to the changes in technology and the organization. Both aspects are being addressed.

Question: What are the main features and core competitiveness of Changan compared to its peers at this stage?

Tao Ji: As an OEM, why does Changan invest so much effort in building self-research capabilities, especially in the AI-driven era? We need to establish such capabilities. We believe that future vehicles will still be complete AI entities, and we will examine them from this perspective.

So how do we define the role of AI in vehicles from a holistic perspective, and how does it work collaboratively? This is a unique advantage of OEMs; only OEMs can fully define every aspect from head to toe, from the brain to the external components, and how this architecture should be aligned.

However, in the past, OEMs lacked the capability to define these aspects. Some suppliers excel in technological innovation, but they inherently find it difficult to design vehicle AI from a holistic perspective. What Changan is doing today is essentially realizing the capability to define the vehicle AI architecture and implement algorithms, unifying these elements so that future vehicles can truly become AI vehicles.

The typical characteristics of this AI vehicle include the AI integration of the entire vehicle, with driving being just one manifestation. It will also achieve personalization, as it can understand human needs through interaction with large models and content, thanks to its memory and long-term memory. We should move in this direction. Everything we are doing today is preparing for this.

In the current implementation of the Tian Shu intelligent driving system, we also hope to introduce some practical features of the Tian Shu large model, such as what we call interactive intelligent driving capabilities, which allow driving to understand a user's open language command set.

For example, when passing through certain tunnels, I might say, "Don't slow down this much; remember my habits." Therefore, we will gradually release such functionalities in the Tian Shu intelligent driving system, ultimately leading to the concept of a fully AI-integrated vehicle.

Question: Is it possible for fuel vehicles to also be equipped with advanced intelligent driving?

Tao Ji: Technically, there is no problem as long as the execution mechanisms are drive-by-wire, and the actual power can meet the requirements of domain control and sensor power. However, internally, the EE architecture of fuel vehicles is certainly different from that of the new generation of electric vehicles. To realize these functions, architectural matching, communication interface matching, and signal matching are required. Whether the cost of matching is worth it is something that should be planned, but specific vehicle matching is still to be determined Users have needs, and we respond accordingly.

Question: Changan announced that it can achieve L3 by 2026. What kind of reasoning led to this judgment on the timeline?

Tao Ji: L3 is essentially about technological progress that allows us to see that previously poorly handled interactive games, including roundabouts, can be resolved through end-to-end methods and the support of large models. This is fundamentally the case.

Because L3 requires redundancy, whether it's steering and braking redundancy or domain control power supply redundancy, it needs to ensure that when the main system fails, it can pull over or stop safely, achieving a very high level of safety and reliability.

I believe that with the emergence of high-performance chips today, those over 500T can run larger-scale end-to-end models and execute large models on the edge, allowing for better handling of the long-tail scenarios or difficult interactive scenarios mentioned earlier. Based on this premise, we see that L3 will emerge within 2026.

Question: In terms of intelligent driving, how does Changan build its technological barriers, and what will be the core competitive advantage in intelligent driving over the next three years?

Tao Ji: In the past, OEMs were more familiar with the supplier model, breaking down a large system into subsystems and handing them over to different suppliers while integrating them themselves. However, as we move towards AI and data-driven, model-driven intelligent driving, you will find that this approach is too difficult to establish. It's also challenging to break down into such subsystems, and the coordination and information transfer between subsystems are not like the past methods.

The primary competition still lies in talent and cognitive enhancement. You need to understand how to play this game in the current era and what kind of people should come to do this. I believe that over the past year, Changan's AI and intelligent driving team has made significant progress in this regard, and the core team is at the forefront compared to any new force brand in Beijing, Shanghai, and Guangzhou.

However, translating the understanding of the leading figures into product competitiveness still has a time window, and I think we can wait patiently.

Returning to specific technologies, OEMs are the first entry point for data. In the future, there may be millions of vehicles equipped with such data, and being able to truly leverage this is very core. It's not just about having a vehicle's code or a white box.

What standards and specifications are used to process this data? What methods are used for label extraction, data balancing, automated cloud-based large model storage, and efficient training, etc.? The capabilities in data and infrastructure are what traditional OEMs have lacked but are the core capabilities needed for the next generation of technology.

With this capability, it's like building a very luxurious kitchen; you can cook any dish you want. This is what I see OEMs need to do, and what Changan needs to do.

Question: As a state-owned enterprise, what advantages does Changan have?

Tao Ji: Changan indeed has a very distinctive cultural imprint. From moving from Shanghai to Nanjing to Chongqing, migrating inland, and transitioning from military industry to micro vehicles to passenger vehicles and then to new energy, each major change in geographical location and events has led to a unique style of Changan. During these turbulent times, the unity, cohesion, execution power, and desire for self-survival that emerged have become relatively unique characteristics of Changan When making decisions and executing them, it is quite resolute and efficient. I heard that when they were scrambling for GPU cards two years ago, the decision-making speed was extremely fast, within just two or three days. If they were to follow the traditional state-owned enterprise process, the investment in fixed assets might need to be planned over a year in advance. However, in the face of urgent events, they can really burst forth.

Changan is very willing to engage with advanced productive forces. The chairman often talks about what Lei Jun discussed with him, and it seems they are recently planning to hold discussions on red and blue army products, which they have actually learned from these advanced enterprises.

We can only say that the controlling shareholder is state-owned assets, but we are fully engaged in market competition. All of us are here, and our salaries are linked to our performance. Moreover, we are the only one in the automotive industry that has survived the transition from military to civilian. Additionally, many central enterprises actually do not dare to implement employee equity incentives, but we, in a fully competitive environment, dare to do so, and the approval process is particularly fast.

We have annual strategic discussions in the first half and second half of the year, reflecting on trends. Furthermore, we are actively undergoing transformation, which we now call the third wave of innovation and entrepreneurship, transitioning from fuel vehicles and electric vehicles to a smart low-carbon mobility technology company.

Q: What kind of acceleration or cost reduction and efficiency improvement methods do upstream suppliers like Tencent provide you?

Tao Ji: In the process of autonomous driving, we need maps, and as we gradually move towards the cloud, we also need maps. One aspect is the qualifications of the mapping companies, which are very important for today's data collection compliance and processing compliance. However, not every enterprise has the Class A surveying and mapping qualifications to do this; we must have a good partner to help us handle all compliance matters.

Another aspect is the automated data labeling in the cloud. The original high-precision mapping capabilities will increasingly shift to the cloud, helping us build our models. The high-precision data is actually the ground truth for our models and the input for our model training.

In this changing process, I believe the Tencent Maps team has provided us with a great transformation, turning map data into model data. This process helps us collaboratively build capabilities from data collection to compliance, model training, and deployment in vehicles. I think this is a new symbiotic and win-win approach between OEMs and partners in the future.

Including some model training and inference in this process, we have also found that Tencent Cloud currently has excellent linear acceleration capabilities for GPU cards, which is indeed one of the reasons we are willing to use Tencent Cloud.

So in this entire process, I believe Tencent Cloud has provided excellent underlying technical support, and Tencent Maps has provided us with data assurance from both the vehicle side and the cloud side, enabling us to quickly iterate our models and allowing us to rapidly utilize the lightweight map data provided by Tencent when we are on the road, solving the problem of prior road recognition during navigation.

Q: Will future intelligent driving be a standard component, a standard component that does not emphasize personalization, or will it be a differentiated personalized component that can achieve considerable accomplishments? Tao Ji: I think it must be the latter. Of course, everyone first needs to produce standard components. Once the end-to-end model is established, each company will converge, becoming more like an average driver, driving steadily and smoothly, making it difficult to drive according to your own ideas. However, this stage is a necessary process.

Moving forward, how to differentiate is key. The interactive intelligent driving proposed by Changan last year will allow voice control of the vehicle, treating it as a "person" that can understand your language. If a car can comprehend your words and make corresponding changes in its driving behavior, it will better meet the user's requirements for a dedicated driver. I believe that the large model's long-term memory of personal preferences and continuous learning capabilities will continuously accumulate such abilities, making the car increasingly aligned with user needs. I think this is an inevitable trend. Therefore, interactive intelligent driving is essentially AI for the entire vehicle; we call it AI defining the car.

Q: Can you introduce the situation of Wutong Car Link? This is your earliest collaboration with Tencent.

Liu Tongyang: Wutong Car Link was established in 2018 as a joint venture incubated by Tencent and Changan. Initially, its positioning was to develop ecosystems related to cloud, maps, and navigation. Around 2022, we transformed into a company that integrates software and hardware, providing the industry with some integrated solutions.

Wutong Car Link, from hardware platforms to systems and applications, is entirely focused on the incubation of AI capabilities and the development of AI products. We look forward to the continuous evolution of these AI products.

Q: What role does Tencent Cloud play in the intelligent transformation of automotive companies like Changan?

Mao Yeping: Changan and Tencent have been cooperating since 2017, reaching a strategic cooperation agreement, and by 2018, we were already collaborating with Tencent Cloud on the entire intelligent and connected aspect.

In terms of technology, both parties have jointly built a digital cloud foundation and developed a complete set of R&D toolchains, allowing us to achieve 24/7 cloud service support. Tencent's strengths in the evolution of its vast internet ecosystem have greatly contributed to the evolution of our mobile app for vehicle control.

Ultimately, people, cars, roads, and clouds need to develop towards an ecosystem. Tencent's subsequent influence and the direction we want to take together require us to continue innovating and collaborating with Tencent.

Q: After the release of the Xiaomi car last year, was there any discussion within Changan?

Mao Yeping: The impact of Lei Jun's several press conferences on the automotive industry, especially regarding the operation of traffic and communication models for automotive companies, was significant—one could say it was like an earthquake. The leadership has also urged us to start making some changes. I believe the automotive industry has gone through two rounds of transformation. The first round was when companies like XPeng and NIO, along with several internet companies entering the automotive sector, had an impact, but it did not shake the foundation; everyone remained stubbornly attached to their old ways After the real ICT comes in, things will be different. It's not just Lei Jun, but also Richard Yu from Huawei. These combinations have led to very in-depth discussions for all of our automotive companies, and the impact is particularly significant.

Q: How strong is Changan's commitment to self-developed intelligent driving?

Mao Yeping: From the very beginning of our focus on intelligence, we set this direction in 2017, and there have been three rounds of entrepreneurship since then. We have cumulatively invested 114.8 billion in enhancing our R&D capabilities. We have gradually built a team of 5,000 people in software and AI, which is now fully formed.

The subsequent series of transformations, including end-to-end investments, are enormous. We are continuously planning this action together with our superior units and partners, with an investment of 200 billion across the entire new vehicle industry chain. The core focus is on intelligent driving and intelligent cabins, and we aim to build a team of 10,000 dedicated to this area within the entire intelligent sector.

Regarding future product layout, our heavy investment is not only focused on R&D but also on product planning. Changan Automobile's current intelligent driving, flying cars, and humanoid robots all share very similar algorithms from perception to regulation to mapping. Therefore, we will also extend our industrial chain layout in the industry.

Our aircraft will take flight by the end of this year, specifically eVTOL, which is directly L3 and cannot be operated manually, enabling point-to-point, end-to-end intelligent flying. Additionally, by 2026, our humanoid robots are also accelerating their progress. Recently, our team has been actively discussing various projects, and things are moving very quickly in this industry.

In total, we plan to invest 50 billion specifically in flying cars. We are also pursuing a dual approach by establishing a joint venture with EHang in Shenzhen to develop flying vehicles. EHang currently has a complete set of domestic licenses and is leading the way, essentially collaborating with the top players in intelligent driving.

Q: In the future, if survival is the goal, does that mean full-stack self-development is necessary?

Tao Ji: I don't think it's that absolute. For each company, some are R&D-oriented and technology-driven, while others are more commerce-oriented and doing quite well.

We can't say we are fully self-developed; we prefer to call it fully controllable. We have our own research and understanding of these technologies, which allows us to communicate effectively with our partners about the optimal direction for Changan and our products. It's not about receiving black boxes from partners and just implementing them.

However, we find that users do not have such demands. When we want to make adjustments, we can't, and if partners say it's not possible and that we don't understand, we certainly lose our voice. Therefore, Changan Automobile has always aimed for full-stack controllability in key technologies, particularly in the fields of new energy and intelligent connected vehicles. But we are not saying we want to take away everyone else's jobs.

Q: What are the future plans for the SDA platform? Mao Yeping: The overall compatibility of the SDA architecture is still very strong, and it is a brand new platform for future smart vehicles. We hope that the SDA platform can be branded and not only empower ourselves but also create value for the industry, so the imagination space for SDA is very large.

The SDA platform is really not just about cars; its central ring network architecture, EE platform, and overall structure should not be narrowly viewed as just a vehicle architecture. Any brand and product can actually utilize our SDA platform. In the past, when new energy vehicles transitioned from fuel vehicles to electric vehicles, they actually modified the architecture of fuel vehicles to create electric vehicles. Now, in this new round, many are using new energy platforms, and SDA is specifically developed for the entire digital and intelligent new vehicle in the future.

From a technical perspective, the evolution of the architecture may involve computing power, communication, information security, functional safety, and a series of actions for ecological integration. Currently, SDA meets the demands of the current situation very well, and of course, there will be a series of evolutions based on this, including the integration of advanced intelligent driving, L4, and the future low-altitude economy.

Q: End-to-end is a relatively certain technical direction now; will there be any new concepts?

Tao Ji: I believe that end-to-end is a relatively certain and necessary step at this stage, and to go further, we must first step on it. Moreover, we have seen some obvious benefits compared to traditional methods, especially in the interaction with dynamic and static obstacles, detours, etc., which are done in a more human-like and smoother manner.

VLM is a more thorough end-to-end approach; it uses a large model to completely connect sensors and final behaviors, including logic and control, in an end-to-end manner. However, given the limitations of the computing power platform in vehicles today and the controllability, it is still quite difficult to move directly to VLM. The reasoning power of large models requires high demands on the edge, and its reasoning frequency involves not only computing power but also bandwidth, which are challenges for today's domain control SOCs, and I think breakthroughs in hardware are still needed.

In the process of reaching the endpoint, we need to continuously improve hardware capabilities, chip capabilities, memory, bandwidth, and other indicators, as well as further rapid evolution of models and algorithm capabilities. The fundamental unchanging factor, in my opinion, is still data; this data is definitely the most important nutrient in the AI era. How to fully acquire this data? How to manipulate it is crucial.

On the way to the end, it may not only involve using today's road-collected data for imitation learning, because today end-to-end essentially uses human driver behavior for imitation learning. The most intuitive analogy is Go; AlphaGo does not need to imitate human chess strategies; it completely develops through self-play reinforcement learning.

Q: In the battlefield of intelligent driving, what kind of competitive landscape do you think Changan Automobile, BYD, and Xiaomi Automobile will present in the future? Tao Ji: I think it's most important to focus on doing ourselves well, including self-research and introducing suppliers, all centered around me, taking the lead, making ourselves good enough to provide consumers with truly useful and loved products as mentioned earlier; the rest will naturally follow.

Question: In this part, the price war in the automotive sector has already been intense. Where will the increased costs of intelligent driving be absorbed?

Tao Ji: Today, Chinese consumers seem to lack the habit of paying for subscriptions. You shouldn't make this a separate visible factor when they are not yet in a state of love for the product. Of course, overall profits are being compressed because if you haven't achieved love for the product, the consumers' willingness to pay for the whole vehicle is not that high, which indeed compresses profit margins. As mentioned earlier, there are some cost-reduction measures and methods for the supply chain that should be utilized.

Mao Yeping: In fact, among all domestic brands, the overall product of independent brands has brought a premium and a willingness to pay in terms of intelligence and new energy. Over the past five years, the average price of 80,000 yuan has basically doubled. This means that consumers recognize the value of this car; what they once thought was worth 80,000 yuan, they might now think is worth over 100,000 yuan. Of course, our higher-end products are sold at even higher prices, which cannot be separated from the support of intelligence and new energy.

Question: Is Changan Automobile currently integrating Tencent's content ecosystem into its intelligent cockpit?

Liu Tongyang: We started integrating more in 2019, such as Tencent's voice, Tencent Maps, and WeChat. Now overseas may be an important battlefield for us. Currently, some of Tencent's ecosystems are also cooperating with Changan, including gradually launching models for overseas markets, which is the second point.

In the future, we may also focus on AI aggregation and capability development. Additionally, we are working on connection management, such as integrating WeChat and linking it with the vehicle's ecosystem. Overall, the scope of cooperation is quite broad and deep, as it is also the parent company of Wutong Car Networking.

Question: What experiences does Changan have in successfully achieving its self-research goals in intelligent driving?

Tao Ji: I think last year we were able to quickly implement functions like highway navigation and parking, which still relied on the industry's development and accumulation over the past few years. Based on BEV perception and AI prediction combined with traditional control methods, this approach is relatively mature and has reached a fairly mature stage. Intelligent driving itself is also entering a phase of large-scale deployment.

At this stage, we are more focused on utilizing the mature experience of skilled talents to quickly engineer and deploy these functions. I think it tests the accuracy of grasping the overall direction and execution capability.

On this basis, we indeed achieved mass production at a relatively fast pace last year. However, we also see that the entire technical paradigm is undergoing drastic changes. Perhaps before the previous generation could complete larger-scale mass production, new technologies have emerged, and we also need to respond to new technical challenges. Therefore, in terms of talent reserves, technical reserves, infrastructure, and data computing power reserves, we are also rapidly preparing Question: From the user's perspective, what they want is a lower-cost intelligent driving system that does not compromise on experience. From your standpoint, how can you ensure this?

Tao Ji: Firstly, from the perspective of cost reduction, I believe that the implementation of universal intelligent driving can increase the shipment volume, and with the increase in shipment volume, the economies of scale in production can help dilute costs. I believe the costs will be more advantageous compared to the previous low-volume shipments.

From the perspective of chips and sensors, we also see that domestic chips and domestic sensor alternatives are gradually maturing, which can help us achieve significant cost advantages.

In terms of these chips and sensors, how can we deploy and optimize algorithms that originally required high computing power? We have done a lot of work in technology and algorithms to ensure that the entire algorithm can run efficiently on a constrained computing platform while maintaining a very good user experience in high-frequency scenarios, which requires a lot of work from the algorithms.

In fact, we anticipate that in the future, once autonomous driving is fully developed, the difficulty of optimization will become easier because there are more mature methods for model optimization and distillation, and it won't require developers to rewrite every rule. We believe the adaptation difficulty will decrease.

For our vehicle models, relatively speaking, our platform-based products can achieve strong universality. As long as the underlying architecture and platform are unified, our intelligent driving solution only needs to supplement a small amount of sensor data for fine-tuning, which I believe can provide a very good experience.

Question: In the future, will it be possible to buy high-level intelligent driving fuel vehicles for around 100,000?

Tao Ji: From a technical standpoint, there are no barriers. The difference between fuel vehicles and electric vehicles is merely the source of driving power; the execution mechanisms can be the same, including drive-by-wire systems for steering, braking, and acceleration. Regarding power supply, I believe that today's fuel vehicles can meet the power supply needs of mid-level computing platforms and sensors, as long as these points are satisfied, there shouldn't be major issues.

However, today the main issue is matching the vehicle model's cost and the price support of the vehicle itself. A typical mid-level intelligent driving system may already be within 5,000 yuan, which is 5% of the cost for a 100,000 yuan vehicle. It depends on how much the car manufacturer is willing to invest in the intelligent features they want to achieve. I personally believe that a 5% cost ratio is completely acceptable. Currently, Changan also has corresponding plans.

Question: Why did the first shot of intelligent driving happen in Chongqing? Because when I chatted with them today, they said it might be due to the complex terrain.

Tao Ji: Chongqing is also one of the cities with the most complex terrain in China, naturally providing the best training ground for intelligent driving. Many intelligent driving companies are live-streaming their 8D magic in Chongqing; we don't need to do that, as we are training in 8D magic every day. It is only right that we showcase the results of our training to everyone as early as possible