Yang Dongsheng, Senior Vice President of BYD: BYD has the confidence for full-stack self-research and will not reduce investment in assisted driving | Exclusive interview by 36Kr

After "popularizing intelligent driving" and "providing parking support", BYD's intelligentization has entered the fast lane.

On July 28th, BYD held a small - scale media communication meeting. Yang Dongsheng, the dean of the New Automotive Technology Research Institute, Han Bing, the deputy dean, and several technology leaders had an exchange with media such as 36Kr.

Regarding the specific effects brought by BYD's series of intelligent actions, Yang Dongsheng said, "Since we provided parking support, we've seen that car owners are using the functions more confidently."

He revealed to 36Kr that currently, the usage rate of the parking function among BYD users has increased from the original 30 - 40% to over 60%. Affected by users' mental states and trust levels, the user stickiness of the urban NOA is only around 20 - 30%. However, the peak usage rate of the highway NOA has once reached 90%.

In February this year, BYD launched 21 models equipped with advanced driver - assistance systems all at once and introduced three platforms, namely the Heavenly Eye A, B, and C, under the BYD driver - assistance system.

The A platform is equipped with dual Orin X chips with a computing power of 508 TOPS and 3 lidars. It is positioned as ultra - high - end and is mainly installed in the Yangwang brand. The B platform is only equipped with 1 Orin X chip with a computing power of 254 TOPS and 1 or 2 lidars, and is mainly installed in the Denza and BYD brands.

The C platform is the entry - level version. It has the largest scale of intelligent driving in BYD and is currently the key self - developed platform. It is not equipped with lidars and uses the NVIDIA Orin N platform and Horizon J6M, and is mainly installed in the BYD brand.

At present, functions such as highway and urban expressway navigation and assisted parking can be realized. BYD said that the urban memory navigation function can be realized by the end of the year.

Since then, from the 249,800 - yuan Song L EV to the 78,800 - yuan Seagull model, all have the ability to achieve advanced driver - assistance.

At the beginning of July, these 21 models received a large - scale OTA upgrade in terms of driver - assistance. In addition to the function upgrade, the core point of this OTA is that BYD announced that it will fully cover the safety and losses of Heavenly Eye users in the intelligent parking scenario.

This means that for accidents that occur to users in the parking scenario, BYD will be responsible for after - sales maintenance and other issues, and users can avoid the risk of increased insurance premiums due to insurance claims.

In terms of continuing to break through in driver - assistance capabilities, BYD believes that power and chassis are its core advantages. The capabilities of the e - Quad and e - Tri platforms can greatly improve the efficiency of driver - assistance, and efficiency will be an important indicator determining the user stickiness of urban NOA.

In terms of the driver - assistance R & D route, Li Feng, who is in charge of intelligent software development, said that the currently popular VLA technology route in the industry is not yet mature. BYD has not invested a large amount of resources in researching the VLA solution and will still mainly focus on the end - to - end and VLM routes.

Yang Dongsheng revealed that BYD currently does not have a clear plan for self - developing high - computing - power chips. He believes that the real challenge of driver - assistance lies in algorithms and data. "We still hope to deploy the algorithm model capabilities on chips with medium and low computing power so that models with 100, 300, and 500 TOPS computing power can also perform well."

He said that BYD will not reduce its investment in self - developing driver - assistance technology. Self - development ability is the prerequisite for in - depth understanding and taking the initiative. However, BYD is large enough. In the future, it will continue to maintain an open attitude, introduce algorithm companies for cooperation, and keep "walking on two legs".

It can be seen that the future development direction of BYD's driver - assistance will still focus on safety and popularization, and will focus more on building user trust by improving the actual experience and reliability rather than simply pursuing the "speed" of technological iteration.

BYD's goal is to make the driver - assistance function truly integrate into users' daily car - using habits and significantly improve user stickiness. Even the owners of models below 100,000 yuan can develop the confidence and habit of using driver - assistance.

The scene of the communication meeting

The following is a partial dialogue between media such as 36Kr and executives such as Yang Dongsheng, slightly edited without changing the original meaning:

Question: What effects has the equalization of driver - assistance brought so far? Is there any data to share?

Yang Dongsheng: Currently, the usage rate of parking has increased from the original 30 - 40% to over 60%, and this was the data I saw a week ago. The usage rate on highways and expressways is the highest, with a peak of 90%.

The user stickiness in urban scenarios is not that high. We believe that the urban scenario depends 100% on users' mental states. Currently, we have two evaluation indicators for urban scenarios, one is safety takeover, and the other is efficiency takeover.

If you don't care about efficiency and are not in a hurry, you can actually use it all the time in the city. So we've seen that the urban data has also gradually increased from 10% to 20 - 30%.

Question: Has the Heavenly Eye C played a role in sales?

Yang Dongsheng: It took us 10 years to popularize electrification, and the popularization of intelligentization may be a bit faster. However, there are still few models with intelligent driving below 100,000 yuan. It takes a certain process to build the awareness of this consumer group, and BYD has also taken on the task of popularizing intelligentization education.

The parking support we announced some time ago has also enhanced users' confidence in using the functions. Trust can change users' usage habits. We are trying to make users get used to the process of using intelligent driving from a safety perspective.

Question: What's the progress of BYD in the VLA large - model technology? When will it be installed in cars? This technology route requires a large amount of computing power, and generally more than 1000 TOPS in the industry. Will BYD self - develop high - computing - power chips?

Li Feng: Currently, the VLA route is not very mature, and the results it produces may not be as good as the end - to - end route. So we are actually not investing much energy in VLA R & D.

Yang Dongsheng: For engineers, the larger the computing power, the better. But we hope to deploy the algorithm model capabilities on chips with medium and low computing power so that models with 100, 300, and 500 TOPS computing power can also perform well. So in BYD's self - development plan, there is currently no clear plan for high - computing - power chips.

Question: How do you view the currently widely - discussed self - developed model of automobile enterprises' full - stack and the supplier model, as well as the future trend?

Yang Dongsheng: We have excellent vehicle - building capabilities and the ability and confidence to conduct full - stack self - development, and we have the initiative. Currently, we have both self - developed and cooperative projects. In the future, we will still introduce these algorithm companies with an open attitude. We think we are large - scale, and our own participation helps us deepen our understanding, and we will adhere to this long - term cooperation method.

Question: BYD believes that the effect of driver - assistance is closely related to data training. So what efforts has BYD made in terms of data, especially high - quality data?

Yang Dongsheng: Currently, in order to make driver - assistance more anthropomorphic, one approach in the industry is to use experienced driver data for training, and we are also improving our capabilities in this regard. If a person has very good driving habits, then 80% of the scenarios may be "experienced driver data". So we have a set of mechanisms to judge driving behaviors to find the "experienced drivers" among our users. We are also constantly improving the application of AI large models in data. We have a dedicated data team within the company.

Question: What technological innovations does BYD have in the integration of intelligent chassis and driver - assistance?

Yang Dongsheng: BYD's strengths lie in the power and chassis parts.

Currently, the capabilities of the e - Tri and e - Quad have been mass - produced and launched in the parking function. The e - Quad is called "shaking head and wagging tail", and the e - Tri is called "dragon wagging its tail". We've taken the first step. The innovation of the chassis and power system has greatly shortened the parking time. The improvement here is not just 5 - 10%, but a 50 - 80% time savings.

At the technology press conference, we also demonstrated the unmanned moose test of the e - Tri and e - Quad, which can make the vehicle more agile when making a U - turn on urban roads. Of course, more verification and testing are needed to ensure safety, and more testing is required before OTA iteration.

Question: Currently, the public has a strong sense of anxiety about the safety of driver - assistance. What technological reserves does BYD have to deal with this situation?

Yang Dongsheng: Currently, the safety coefficients of parking and highway/expressway navigation are relatively high, but the driving scenarios are very complex. The traffic lights and the situation of electric vehicles in each city in China are different, and some working conditions are extreme, such as the sudden appearance of electric vehicles. The technical challenges are very great.

However, some safety accidents are not entirely determined by technology. The tire grip determines the ultimate safety of the vehicle. So I think there is a market for tires and positioning. We need to keep finding ways to improve tires, and at the same time, rely on the power chassis to improve the vehicle's response ability.

In the era of fuel vehicles, it was very difficult for us to accurately calculate the slip between the tires and the ground, so it was also difficult to adjust the braking distance. However, the electric motor provides such an opportunity. Now we can accurately calculate how much power is needed based on the amount of slip between the tires and the ground, and then output the right amount of power. In this way, the safety of the vehicle will be greatly improved.

What we are currently doing in the combination of electrification and intelligentization is exactly this kind of thing.

The scene of the communication meeting