Decrypting BYD’s Smart Driving Strategy

Behind BYD's self-development of all-in-one smart driving technology, there are not only flowers and applause, but also thorns on the ground.

BYD's intelligent driving technology capabilities seem to have improved exponentially overnight.

Although founder Wang Chuanfu once publicly stated: driverless driving is nonsense. But he also admitted that high-level assisted driving is practical. And BYD is "speeding to catch up" on this new track.

Recently, BYD has been "crazily" releasing the progress of its intelligent business on various occasions. It seems that it wants to remove the "unintelligent" label overnight, and both electric and intelligent products must enter the first echelon.

Among them, at the functional level, from the L2-level Dipilot driving assistance system purchased from Bosch, which has been criticized in the past, to the self-developed high-speed DNP (high-speed pilot) high-end intelligent driving assistance, related functions will be launched in the third quarter of this year. boarding.

At the hardware level, BYD's self-developed domain controller has a computing power of 508TOPS. If nothing else, it should be the mainstream dual OrinX combination in the industry. It will also be launched this year, and the target model is to look up to the U8.

On the technical path, BYD, the industry's leading BEV sensing large model, will also be deployed within this year, and will plan to deploy and occupy the network in the next step.

Δ Han Bing reveals the latest progress of BYD’s intelligent driving business

It seems that what Tesla, Xpeng, etc. took two or three years to achieve, BYD will have to catch up with in a short time. But this is a phenomenon that violates normal technological understanding. The research and development of intelligent driving has never been something that can achieve results in a short time by stacking people and materials, and BYD is no exception.

In fact, after sorting out BYD's past layout in the field of autonomous driving, we found that its accumulation in related fields has also gone through the process of cognitive establishment, cooperative research and development, and then the formation of a global self-research strategy.

To a certain extent, high-speed DNP can be seen as the first step in BYD's long march of intelligent driving. It is the result of initially establishing self-research capabilities through multi-party cooperation. Although it cannot catch up with mainstream players, it can be regarded as reluctantly responding to the "BYD There is a long-standing stereotype of "unintelligent".

But the follow-up story of BYD's global self-research is not only full of flowers, but also full of thorns.

01

The beginning and end of the layout, a long-distance race lasting 8 years

As Han Bing, director of BYD's electronic integration department, replaced Wang Huan and took over the position of head of BYD's intelligent driving business, he also officially became the de facto leader of BYD's intelligent driving software and hardware R&D team.

In the short period since then, Han Bing has frequently appeared in various public events, and BYD's latest developments and research and development results in intelligent driving have also been made clear to the outside world through Han Bing's mouth. To briefly summarize, there are probably a few major things worth paying attention to:

1. BYD's high-speed DNP (high-speed pilot assisted driving) will be launched in the third quarter of this year. The first model will be BYD Han. This solution will be based on the Horizon J5 chip. The sensor is a 11V5R de-lidar combination. The domain control is provided by Neusoft. Provided by Ruichi, the software algorithm is self-developed. In addition, this solution also includes high-precision positioning services provided by Juefei Technology. It is most likely a solution that relies on high-precision maps;

2. At the technical level, BYD has internally developed a BEV perception model with multi-camera fusion, which can be purely data-driven and at the data accumulation level. It has accumulated more than 150PB of data, and 1PB of data will be added every day. The data is annotated. The automation rate reaches 95%;

3. At the hardware level, BYD’s self-developed large computing power domain control solution will soon be installed on vehicles, and the target vehicle is  the U8  (  parameters  |  inquiry  ). It is reported that the computing power of this set of domain controllers is 508TOPS, and it will not be surprising that it will be a combination of dual OrinX.

4. In the future, BYD will conduct a large data-driven model for the entire process of sensing, prediction, and decision-making planning (currently, except for sensing, all are rule-driven) and deploy occupied networks.

…

It’s hard to imagine that BYD seemed to have shed its “unintelligent” hat overnight and stood at the forefront of intelligent driving technology.

But this is obviously not the case.

In fact, if we look back on history, although BYD has a relatively low presence in intelligent driving, its related layout has never stopped. Tracing back to the starting point, it first began with cooperation with Baidu in autonomous driving.

In October 2015, BYD had already reached a cooperation agreement with Baidu on autonomous driving technology before the establishment of Baidu's autonomous driving division. After the establishment of Baidu's autonomous driving division, Wang Chuanfu also served as its autonomous driving consultant. In October 2016, BYD began delivering modified driverless test vehicles to Baidu.

Three years later, BYD and Baidu once again disclosed that Baidu will provide BYD with an L3 level autonomous driving solution, and the two parties will achieve mass production of autonomous vehicles within another three years.

It can be said that from the beginning of Baidu’s autonomous driving business layout, BYD has been its staunch partner and runner.

But since then, no substantive news has been released about the cooperation between the two parties based on high-end autonomous driving. The most recent one was in February last year. It was reported that Baidu would provide BYD with ANP smart driving products that integrate parking and parking. There is a map for shared driving, but there is no clear progress afterwards.

More voices are saying that BYD has begun to change direction and plan for lower-level intelligent driving, focusing on both investment and business cooperation.

In the first half of 2021, BYD invested twice in AI chip manufacturer Horizon, and also invested in lidar supplier Sagitar Juchuang during the same period.

In terms of business cooperation, at the end of 2021, BYD and autonomous driving software solution provider Momenta established a joint venture company Dipai Zhixing, with BYD accounting for 60% of the equity. The two parties will seek cooperation in high-end intelligent driving. Interestingly, Han Bing’s title in recent public events is not only the person in charge of BYD’s intelligent driving, but also the CTO of Dipai Zhixing.

Soon after, BYD and NVIDIA and Horizon came to the news about AI chips, which should be the high-speed DNP and BYD's self-developed large computing power domain control platform.

Actively deploy externally and recruit people frantically internally. Starting in 2022, BYD’s school recruitment will begin to see a large number of job demands related to intelligent driving such as autonomous driving control and visual algorithms, and the academic requirements are not low, with PhDs or postdocs being the main ones. This phenomenon will be revealed in BYD’s autumn recruitment this year. There is still no change, and there is a strong momentum of "all the heroes in the world are among my heroes".

Therefore, BYD Executive Vice President Li Ke said at an internal investor communication meeting that BYD has currently recruited a software team of four to five thousand people in autonomous driving, which may not be false.

If you just purchase other companies' solutions and do the engineering on your own, you don't need such a huge software team. But if you do self-research, that's a different matter.

02

Drive wisely in three steps, and the path will gradually become clearer

It can be seen from the above-mentioned independent incidents of Zero Zero Total that although BYD is relatively low-key in terms of intelligent driving, its path is already relatively clear.

To put it simply, it can be roughly divided into three stages: establishing awareness, cooperation and reference, and then establishing full-stack self-research.

The so-called establishment of cognition is actually the correction from autonomous driving to intelligent driving. Judging from BYD's accompanying Baidu autonomous driving, at least at the beginning, BYD's understanding of autonomous driving is by no means "the emperor's new clothes" as Wang Chuanfu said.

However, it was not until the second half of 2021 that the entire industry's attitude towards autonomous driving changed fundamentally. The leap-forward high-end autonomous driving route was basically denied. The prospects of robotaxi companies including Baidu, Pony.ai and other robotaxi companies were questioned. Car companies such as Tesla, Xpeng and Momenta have been corrected on their progressive routes.

So after 2021, BYD began to enter the second stage, cooperating and drawing on the foundation of self-research capabilities.

At least from the perspective of layout, BYD's prospects for high-end autonomous driving were in line with the market trend during the same period or even earlier. It was also at this time that BYD began to intensively deploy intelligent driving. Join forces with Momenta, Horizon, etc. to seek excellent collaborators in the industry and learn from them.

In fact, moving from cooperation to self-research is a common path for traditional car companies to lay out intelligent driving. Among the many things BYD will do in 2022, it will look for partners externally and expand its team internally to complete the transition from cooperation and reference to learning and self-research. Prepare and come up with a preliminary result: high-speed DNP.