In the era of big models, the three AI giants SenseTime, Megvii, and iFlytek are competing for intelligent driving

The competition in the intelligent driving market in 2024 is already in a convergence phase, but there are still new giants entering this extremely involutionary track. They are AI platform companies that excel in algorithms, such as SenseTime, Megvii, and iFlytek.

At the China Electric Vehicle 100 Forum, He Xiaopeng, chairman of Xpeng Motors, said that the last decade was the decade of new energy, and the next decade will be the decade of intelligence.

The level of automotive intelligence is becoming more and more attractive to consumers, and has become a differentiated capability that car companies are working hard to build. In the face of this trend, in order to maintain their leading position in the industry, car companies have built core competitiveness in intelligence through self-research and acquisitions. Emerging companies represented by Horizon Robotics have been working hard for many years and are finally reaping the rewards. Traditional Tier 1 giants are also unwilling to lag behind and are accelerating the layout of intelligent core technologies.

Now, new forces are beginning to rise. These companies have been dormant for a long time, but due to the uncertainty of the market, they are just conservatively following or doing some pre-research.

For example, SenseTime started participating in Honda's autonomous driving project as early as 2016, when the market was generally enthusiastic and expectant about L4 intelligent driving. However, in the following years, apart from some demonstration projects, there were few commercial projects.

At the beginning of 2021, things changed.

• 
  

• SenseTime established the Intelligent Vehicle Business Group, sending a stronger signal to the outside world;

• 
  

• In the same year, Megvii also promoted smart driving as a key business;

• 
  

• In the first half of 2022, iFlytek’s intelligent driving subsidiary surfaced.

1. Why this time point?

Technically speaking, the AI ​​algorithms that these companies are good at have always been the core of intelligent driving, especially the computer vision that perception relies on. The founding teams of each company also have a strong academic background.

But before this, a major challenge faced by pure AI companies entering the field of smart driving was the diversity of hardware platforms, including different chip platforms and sensor categories.

This makes it difficult for AI companies that excel in algorithms to use one set of software to adapt to various hardware. It also forces them to run around in different projects with customized requirements, and the security requirements are often not met.

In addition, cost is also an important factor.

At the beginning, in order to ensure driving safety, the route of intelligent driving is multi-sensor fusion and configuration of high-computing chips, which is difficult to meet the low-computing and low-cost requirements of car manufacturers. It should be noted that the operating profit margin of traditional car manufacturers is generally between 5% and 10%.

It was not until 2020 that Tesla first proposed the BEV algorithm and successfully implemented the NOA function using data from 8 cameras, verifying the intelligent driving of pure visual routes and making costs controllable.

Guo Jishun, vice president of Joyson Electronics, once said in HiEV's live broadcast column that after the birth of BEV, many companies began to propose training various sensors based on BEV, and then obtained a BEV platform, on which they can adapt more cameras with different pixels, cameras with different angles, lidars with different principles, or 3D, 4D millimeter-wave radars.

This new paradigm greatly reduces the obstacles that hardware changes bring to software and algorithms, which gives AI companies more opportunities. After all, they are better at intelligent algorithms.

2. SenseTime Jueying: High-profile and high-profile to create a flywheel effect

SenseTime started its layout in intelligent driving early on.

In 2016, when the autonomous driving craze was at its peak, SenseTime began to participate in Honda's autonomous driving project. The two companies signed a long-term cooperation agreement in 2017 to jointly develop L4 autonomous driving technology, with SenseTime providing visual perception technology as well as chips and embedded systems for autonomous driving. Since then, autonomous driving test bases have been established in Joso City, Japan, Hangzhou, China, and Lingang, Shanghai, to demonstrate autonomous driving on semi-open roads.

SenseTime publishes many papers on autonomous driving at academic conferences such as CVPR and ICCV every year, and has won many championships in academic competitions related to intelligent driving. In 2023, it won the CVPR Best Paper Award for its paper on the UniAD end-to-end autonomous driving large model.

At the World Artificial Intelligence Conference in September 2019, SenseTime released the "SenseTime Intelligent Driving Mass Production Solution" and the "SenseTime Technology Autonomous Driving Solution."

The first model equipped with SenseTime's ADAS advanced driver assistance solution is Nezha V, the third mass-produced car of Nezha Auto under Hozon.

However, the ADAS solution at this time mainly provides warning functions, such as lane departure warning, front vehicle collision warning, etc., and has not yet achieved lateral or longitudinal control of the vehicle. It can be said that in the past few years, SenseTime's automotive business has been mainly focused on smart cabins, and smart driving has been mainly explored, lacking commercial cases.

At the 2021 WAIC World Artificial Intelligence Conference, SenseTime Jueying was officially launched. Subsequently, SenseTime established the Intelligent Vehicle Business Group, and the status of its intelligent vehicle business has risen sharply.

SenseTime’s automotive business is led by Wang Xiaogang, one of the company’s co-founders, and the team size is over 1,000 people.

SenseTime has always been aggressive in the field of AI, and its automotive business is no exception. Its main products include vehicle cabins, mass-produced intelligent driving, autonomous driving functional vehicles, vehicle-road collaboration, basic platform engines, etc.

Although this strategy requires a large investment in R&D and infrastructure, it can enable the business to obtain more support from cutting-edge technologies, which is crucial, especially in the era of large models. In addition, diversified business needs can dilute R&D costs.

Faced with the rapidly growing market of intelligent driving, SenseTime hopes to create its own flywheel effect in this way.

This is only beginning to show up in the financial data.

According to SenseTime's 2023 annual financial report, the automotive business revenue is 383.7 million yuan, of which the growth of smart driving-related revenue is indispensable. Although the automotive business only accounts for 11.3% of the total revenue, it has increased by 31.1% year-on-year, which is still a good growth compared with the traditional AI business.

Currently, SenseTime has mass-produced high-speed NOA for models such as GAC Aion LX Plus, Nezha S, and GAC Haobo GT, and will achieve mass production of urban NOA this year.

Among them, the intelligent driving function of Nezha S has been officially installed on the vehicle through OTA upgrade in February this year, based on the 16 TOPS computing power domain control platform and 5R11V sensor configuration.

The Nezha S project is a high-speed NOA solution based on dual TI TDA4 VMs. The TI TDA4 series chips are cost-effective, but have low computing power. On this basis, mass production of stable high-speed NOA functions poses a strong engineering challenge, and there are not many manufacturers in the market that can do this.

In addition to the ANP2 team led by DJI and Baidu Su Tan, SenseTime is currently the only team that has mass-produced this solution. This is also a great recognition for SenseTime's engineering capabilities in addition to its algorithm capabilities.

At present, SenseTime’s market value is around HK$20 billion, which is lower than the valuation of some unlisted autonomous driving companies. Regardless of who is overvalued or undervalued, the smart driving business will definitely bring good performance bonus to SenseTime and have a positive significance for the increase of market value.

3. Megvii: Pursuing the Ultimate Cost-Effectiveness

In November 2018, Megvii publicly demonstrated its in-vehicle AI vision solution, which was also mainly in the cabin, such as face unlocking, account switching, driver recognition, multi-modal interaction and other functions based on the in-vehicle system and driving process. In the prospectus published in 2021, there is not much information related to intelligent driving.

It was not until 2023 that Megvii revealed the latest progress in the field of intelligent driving.

According to previous interviews, Megvii believes that traditional algorithms based on radar + vision cannot solve high-level intelligent driving problems, and its own advantages are concentrated in computer vision algorithms, so it has been waiting for a breakthrough in technology. This situation was broken after the successful verification of the BEV algorithm.

So starting from the beginning of 2021, Megvii officially focused on developing intelligent driving solutions.

At present, Megvii has an independent intelligent driving team of about several hundred people, two-thirds of whom are R&D personnel. The main person in charge is Liu Wei, who was formerly the global vice president of SAP and the general manager of Microsoft's channel division.