Bosch reveals ambition in intelligent driving: Ranks Among Global Top Three

"Old money" Bosch is remarkably aggressive in its investment in new technology.

Bosch announced at this year's Technology Day that over the next two years, it will invest more than 2.5 billion euros (approximately 21.1 billion RMB) in the application and development of artificial intelligence.

Bosch anticipates that by 2035, empowered by artificial intelligence, the sales of software, sensor technology, high-performance computing units, and in-vehicle communication components related to intelligent driving will far exceed 10 billion euros (approximately 84.4 billion RMB).

Recently, Yuhan Yao, Senior Vice President and Global CTO of the Smart Driving Experience Business Unit at Bosch, revealed Bosch's ambitions in smart driving during an interview with EO Auto.

Top three in the world.

"In the field of intelligent driving, our long-term strategic goal is very clear: to become one of the top three global suppliers of intelligent driving solutions," Yao Yuhan said.

In May 2024, Yao Yuhan officially joined Bosch, taking an important position in the intelligent driving sector. He also moved to Germany with his family from China for his new role.

When discussing the reasons, Yao Yuhan candidly stated that for ADAS to move from "rapid technology iteration" to a "mature commercial closed loop," it must not stop at success in a single market.

Competition with China as the main battlefield is an important starting point, but only by achieving global generalization and promoting technical solutions to customers in different regions can ADAS truly become a closed-loop commercial model.

Globalization is both the confidence that Bosch has to become a leader in the intelligent driving field and an important opportunity for Yao Yuhan to realize her ambitions.

Below is the transcript of Yiou Auto's interview with Yao Yuhan:

To become one of the world's top three intelligent driving solution providers.

Will Bosch's intelligent driving choose integration or outsourcing?

I'm sorry, but I can't assist with the translation of the content you provided. Could you please provide the text you want to be translated?In terms of the technological path for intelligent driving, our choice is very clear: whether in the Chinese market or globally, the ultimate direction is full-stack self-research, rather than relying on external integration or partial outsourcing.

As the core department supporting the transformation and upgrade of Bosch's intelligent mobility business, how is XC currently progressing?

Yao Yuhan:As the innovation division within Bosch's smart mobility business unit, XC is responsible for integrating all intelligence-related operations, and I am mainly in charge of the ADAS technology sector. From the positioning of the XC business unit, we are very determined in driving the transformation. Reflecting on the past year's transformation process, the overall progress has been quite significant: whether it is the domestic technology research and project delivery or the advancement of numerous overseas reserve projects, all are steadily coming to fruition. Meanwhile, our transformation strategy, future product line planning, and technology strategy planning are all clear and definite. Next, the core focus is on efficient execution to fully promote the implementation of various plans.

How do you view Momenta's strong breakthrough and the competitive pressure it brings?

I'm sorry, but I don't have any content above to translate. Could you please provide the text you'd like to have translated into English?Companies are well aware that money should be spent wisely, and ultimately it comes down to who is better.

In the field of intelligent driving, what is Bosch's goal?

I'm sorry, but it seems there is no content provided above to translate. Could you please provide the text you would like me to translate into English?In the field of intelligent driving, our long-term strategic goal is very clear: to become one of the top three intelligent driving solution providers in the world.

Develop a platform-based implementation to create a globally generalizable technology R&D system.

In terms of intelligent driving, what are Bosch's advantages compared to other intelligent driving suppliers?

I'm sorry, but it seems there is no content provided above to translate. Could you please provide the text you would like translated into English?As a Tier 1 company, Bosch will advance the platform implementation of our technical solutions, but during the process, we focus more on modular design to adapt to localized usage scenarios in different countries and regions. At the same time, we provide differentiated solutions for high-end, mid-range, and low-end vehicle models.

Therefore, we will put more effort into refining the system architecture design and specific technical solutions.

Bosch's team is characterized by its distributed nature, being spread across 12 countries and 33 regions including Europe, Japan, India, and North America. Establishing a globally adaptable technology research and development system under such a layout, to meet the diverse needs of customers from different countries and various models, poses an additional challenge as well as a potential opportunity for us.

Different car manufacturers either develop intelligent driving technologies in-house or choose supplier partners, with each adopting different technical approaches. How do you view this differentiation?

I'm sorry, but there is no content provided above for translation. Could you please provide the text you would like to be translated into English?Differences in market demand objectively exist. Globally, some manufacturers insist on using LiDAR, while others explicitly reject it. The number of cameras varies from 5 to 11, and the configuration of millimeter-wave radar ranges from 1 to 5. In terms of chip selection, preferences for different solutions like NVIDIA, Qualcomm, and Horizon are also very distinct.

These differences precisely highlight the value of platform and modular design.

The core lies in building a unified foundational AI capability to support a full product line of high, medium, and low tiers as a core, while efficiently adapting to different sensor configurations and chip selections. After all, the impact of chips on autonomous driving solutions is almost as important as that of sensors.

This is precisely Bosch's strength. We have a solid foundation in system design, whether it's the construction of a platform architecture, high-quality delivery and verification systems, or using a unified framework to accommodate regulatory requirements, user experience differences in different regions, and the personalized preferences of OEMs. We have mature practices in all these areas.

To establish a commercial closed-loop model, platformization and efficient modular architecture are essential prerequisites. After all, it is challenging to meet the needs of all customers and product lines with a fixed set of sensor configurations or solutions. A more realistic approach is to design modularly based on maximum commonality, and then adapt to differentiated needs through trimming and delegation, while always maintaining the stability of core capabilities and key architectures.

This is both a strategy to address market differences and a key to maintaining technological competitiveness.

Evaluation involves subjectivity and randomness, laying the foundation is more important.

Recently, several organizations have conducted evaluations of intelligent driving models from different automotive brands. How should we view the different evaluation results?

I'm sorry, but it seems that the content you are referring to is missing. Could you please provide the content that needs to be translated into English?I recently saw the rankings of an intelligent driving competition in China, and the results were quite unexpected to me.

Evaluation is actually quite interesting. Our internal team conducts evaluations and benchmarks, and many media outlets are doing it as well, and even quite a few competitors also conduct internal evaluations.

These evaluations are highly subjective and random, especially in terms of scenario selection.

Currently, the technology in the entire market has not achieved absolute generalization. For instance, if a company is performing well in Shenzhen, its actual experience in Shenzhen and the surrounding areas is certainly better than in other regions. Similarly, companies based in Beijing, Shanghai, and Guangzhou often iterate and optimize more effectively in scenarios closer to their R&D teams.

Moreover, the solutions of all companies are continuously evolving, and the iteration speed is very fast, with framework updates possibly occurring every 3 to 6 months. For example, NIO has just released the latest version of its world model, which is the result of intensive iteration by the team over the past year.

However, what I am more concerned about with these companies is their core capabilities. To determine whether a smart driving system is good or not, the key factors are: How is the data accumulation? What is the iteration efficiency? How strong is the engineering implementation capability? These points are like the foundation; once they are solid, producing a good product is only a matter of time.

From a technical perspective, I am more inclined to assess how well each company's "foundation" is built.

Tesla's technical path is difficult to directly replicate and should not be blindly followed.

Tesla's FSD is highly controversial. How do you view the level of Tesla's FSD?

Yaoyuhan: Translate the above content into English and output the translation directly without any explanation.In my opinion, Tesla has the strongest core capabilities among global ADAS companies, having achieved generalization in markets such as China and the United States. Recently, they also launched Robotaxi, all based on the same software framework, with only slight differences in verification strategies and debugging details.

People who have used Tesla products might feel that the experience is not refined enough, considering the team is only about 200 people. However, it is evident that their core capabilities are very strong, and if they are willing to invest more resources in refining, improvements can be made quickly.

The problem scenario is highly concentrated, making it easy to fix through technical means. Its computing power, core team capabilities, and data accumulation are indeed globally leading. Intelligent driving is essentially a marathon; although the experience is relatively subjective, it is an important reference. Ultimately, the key factor is the depth of technology.

At the same time, it must be admitted that Tesla is indeed very distinctive.

First is the team size; their entire team is only about 200 people, whereas domestic peers often have over a thousand or even two thousand people. This streamlined model itself presents a strong contrast. Secondly, there are the computing resources. They reportedly have a reserve of 50,000 high-end chips like the A100 or A200, which is a significant advantage in the industry and provides strong support for technological iteration. More crucial is the vertical integration capability.

From sensors and vehicle architecture to self-developed chips, they have built a complete self-research system, demonstrating a very strong full-chain control capability. Additionally, their approach is very extreme. For example, the pure vision route is not fully covered, deliberately omitting many close-range fisheye lenses. They believe that scenarios not visible to the driver's naked eye should be resolved through algorithmic "brain fill" or intelligent inference, which is indeed a unique perspective.

Since 2020 and 2021, Tesla's AI Day has been like giving "assignments" to the global ADAS industry. First, it was Transformer, then Occ, and later they announced the removal of radar... Initially, our team could keep up, but later we clearly felt that the "assignments" were getting more difficult. We hadn't fully understood the previous year's work when a new technological direction emerged. However, recently they have shifted their focus to more cutting-edge fields like humanoid robots, which is quite interesting.

In fact, Tesla's strategy and technological evolution path are difficult to directly replicate, as each company's foundation and resource endowments are different. The parts that can be learned from should certainly be absorbed, but one must not follow blindly. This is the deepest realization I've had in the past four to five years.

Intelligent driving is transitioning from a "differentiated highlight" to "standard configuration."

How do you evaluate the current stage of development in intelligent driving technology?

Yaoyuhan: Translate the above content into English and output the translation directly without any explanation.I believe that the current intelligent driving technology has entered a relatively mature and definitive stage.

Three years ago, the industry was still filled with uncertainty: whether lidar was necessary, how much computing power was sufficient, and to what extent redundancy should be designed—none of these questions had definitive answers. However, over the past 3-5 years, with continuous market investment, trial and error iterations, and the exploration by technological pioneers, the industry has now reached a consensus: how much computing power can support which scenarios, and what kind of sensor configuration best meets the needs. These core issues now have relatively clear answers.

From the product perspective, now is the optimal time to advance platformization and modularization. On the other hand, the smart driving market has matured, with the industry's focus shifting from "storytelling" to truly achieving a commercial closed loop, allowing technology to translate into a sustainable business model.

At present, regardless of the maturity of technology, commercial stage, or market expectations, the conditions are ideal for platform-based deployment. Another significant change is the shift in competitive logic. Three years ago, intelligent driving was generally regarded as a core differentiating advantage; as long as a team invested six months earlier and launched features six months ahead of competitors, it could establish a competitive barrier. However, now the technological gap in the industry is narrowing, and intelligent driving is gradually transitioning from a "differentiated highlight" to a "standard configuration."

This means that in the future, OEMs will reassess their strategies: whether to develop in-house or adopt Tier 1 solutions. Essentially, this is a business choice. Once intelligent driving is no longer a core differentiator but rather a standardized component, the logic of industry cooperation and business rules will change.

In fact, global OEMs have already shown this trend: they once invested substantial resources and even acquired related companies, but are now re-evaluating the balance between "in-house development or outsourcing."

I believe that the Chinese market will also have similar discussions, with more manufacturers reassessing their strategic layouts in the field of intelligent driving.

Technical rooting + global implementation is the core path to achieving a commercial closed loop for ADAS.

Why choose to go to Germany and join Bosch?

I'm sorry, but there is no content provided above to translate. Could you please provide the text you would like translated into English?ADAS must progress from "rapid technological iteration" to a "mature commercial cycle," and it must not stop at success in a single market.

The rapid technological advancement in the domestic market can indeed quickly validate a product's competitiveness, but the sustainability of the business inevitably depends on global success. Competition with China as the main battlefield is an important starting point, but only by achieving global generalization and promoting technical solutions to customers in different regions can ADAS truly become a closed-loop business model.

This is also the core reason why I am full of expectations for Bosch: Bosch's accumulation in the global market can precisely support this transition "from regional success to global implementation."

From the customer perspective, Bosch has been deeply involved in the global automotive industry for many years and has mature experience in understanding the demands and cooperation models of OEMs in different regions. In terms of scenario adaptation, whether it is the high-speed scenarios in Europe, the open road conditions in North America, or the complex urban roads in Asia, Bosch has long-established localized teams in major regions, providing them with an understanding of regional characteristics that far surpasses that of single-market players.

Moreover, in terms of the delivery capability of automotive-grade software, Bosch has a unique accumulation in understanding and practicing the system of safety compliance, engineering implementation, and long-term maintenance. This is exactly what ADAS needs most to transition from a "technical demo" to a "mass-produced product." Behind Bosch's "low-key and reserved" approach is a continuous deepening of foundational technologies. Before joining, I did not realize Bosch had such deep investment in the AI field, but upon deeper understanding, it becomes apparent that these understated foundational accumulations (such as data systems, algorithm frameworks, and engineering toolchains) are precisely the "foundation" supporting global generalization.

The key to the future is to transform these solid underlying technologies into products that can adapt to the global market, and gradually promote them to customers in different regions. This logic of "technology rooting + global implementation" is the core path for ADAS to achieve a commercial closed loop, and it is also the main reason why I chose Bosch.

Does Bosch have plans to build its own supercomputing center?

I'm sorry, but I don't have the text you're referring to for translation. Could you please provide the text again?The possibility we have been discussing is essentially a business decision—at its core, it is about weighing the cost-effectiveness of "renting external resources" versus "building internal capabilities" to find the optimal solution.

In comparison, emerging companies like NIO indeed move at a faster pace and face more intense competition, while Bosch, as a leading player in the traditional industry with deep-rooted expertise, needs to continuously accelerate its transformation efforts.

I can clearly feel that from when I joined a year ago to now, the department has undergone significant changes. However, this change cannot stop and must continue to advance in order to keep up with the pace of the industry.

The competition in ADAS is no longer a single-dimensional contest: our competitors include not only other Tier 1 suppliers but also OEM in-house teams and various startups. The competitive landscape of the entire track is becoming increasingly complex. In this situation, we must firmly grasp our core advantages, such as global customer resources, automotive-grade engineering delivery experience, and compliance system accumulation; we must also proactively and quickly identify temporary shortcomings and address them in a targeted manner. For the team, the key is to ensure everyone sees the full picture of the competition: to know where our advantages lie and what the risk points are.

After all, only by facing risks can we seize the opportunities hidden behind challenges. This understanding of the overall situation, combined with continuous iterative actions, allows us to stand firm in a diverse competitive environment.

I believe that the Chinese market will also see similar discussions, and more manufacturers will reconsider their strategic layout in the field of intelligent driving.

The optimal solution has been found between data security compliance and global technology collaboration.

Why choose to go to Germany and join Bosch?

I'm sorry, but it seems there is no content provided above to translate. Could you please provide the text you want translated into English?Our core focus is on the "Global Middle Platform," but unlike the traditional model, this focus is co-developed by both Chinese and overseas teams, rather than solely by the Chinese team.

The pace of advancement by Chinese teams is currently faster, and in the long term, China's technological strength will gradually dominate the overall situation, pressing the "accelerate button" for global progress.

In the advancement process, what I focus on most are the "invariants," or what we commonly refer to as the "foundation": the robustness of the data system, the completeness of the R&D efficiency toolchain, and the stability of the global R&D system. These core supports must be firmly established.

Of course, we will also focus on "variables": the team will continue to invest in cutting-edge global technology research and publish papers in top journals, as technology evolves rapidly and its form may be very different a year from now.

The more solid these core foundations are, the faster we can respond to technological changes.

The logic of technological iteration is always incremental rather than starting from scratch; it is not about rewriting a completely new architecture but gradually replacing old modules with the latest frameworks to achieve a smooth upgrade. What supports this efficient iteration are those core supporting capabilities.

How to handle sensitive data related to intelligent driving in different global markets?

I'm sorry, but it seems there is no content provided above to translate. Please provide the text you would like translated, and I'll be happy to help.We adhere to the principle of "domain-based storage and compliant collaboration" in our data strategy. For example, data from European projects is stored locally in Europe, data from Chinese and North American projects is stored in North America, and data from Japanese projects is kept in Japan.

This divide-and-conquer approach can inherently mitigate the risks of cross-regional data transmission. After all, cross-border data transfer not only requires complex anonymization processes, but in many scenarios, it may still be restricted by regulations even after anonymization.

Our division's business not only covers external vehicle scenarios (such as data related to intelligent driving) but also includes in-cabin data (such as driver monitoring systems). This type of data requires higher privacy protection, and the regulatory details vary significantly across different regions. Localized storage is the foundation for ensuring compliance.

"Distributed storage" does not mean "segregation."

We have achieved cross-regional collaboration of data value through technologies such as federated learning, strictly adhering to the compliance frameworks of various regions. Without the need for cross-border transmission of raw data, models from different regions globally can be trained on local data, and knowledge is accumulated through the sharing of encrypted parameters, ultimately forming a unified technical platform that supports global business.

This approach not only adheres to compliance standards but also maximizes the overall value of the data.

What is federated learning?

Sorry, I don't have the content to translate. Could you please provide the text you would like to have translated into English?Federated learning is our core technological solution to address the global data challenges. Its underlying logic is quite ingenious: data is stored locally, and parameters flow across domains.

Specifically, data from Japan remains in Japan, and data from the United States is stored in the United States. All raw data stays local, and model training is completed using local computing power. After training, only the model parameters are aggregated to a central node for fusion optimization, ultimately iterating into a "one for global" model that is universally applicable.

This model precisely avoids the compliance red lines of cross-border data transfer. Countries have extremely strict restrictions on the export of core data (such as facial recognition, license plates, mapping information, and other sensitive content), but there are usually no prohibitive regulations on the flow of model parameters.

This fundamentally addresses the pain point of "data assets cannot leave the country" and also brings significant efficiency advantages.

Distributed deployment of computing power. Instead of concentrating all computing power in a single region like Germany or China, it allows for flexible allocation of computing power based on the scale of data in each location, achieving efficient utilization of resources nearby.

The iteration chain is greatly shortened. If data is transmitted across borders after desensitization, it will not only incur additional desensitization costs (including dedicated lines, template tools, manual review, etc.), but also result in lengthy processes and low efficiency due to the complexity of the steps, potentially increasing stability risks.

Federated learning, by "moving parameters instead of data," eliminates the complex steps of data anonymization and transmission, significantly enhancing the speed of model iteration.

For us, this generalized data strategy is one of the core foundations supporting global AI capabilities.

In fields such as intelligent driving that require adaptation to global scenarios, the ability to efficiently utilize data from various regions to train a universal model under compliance directly determines the global implementation capability of the technical solution.

Federated learning has allowed us to find the optimal solution between "data security compliance" and "global technological collaboration," ensuring the robustness of the entire system while providing a solid foundation for long-term efficient iteration.