Momenta is now the fastest growing self-driving car company in China

Jia Haonan from Aofei Temple
Quantum Bit Report | Public Account QbitAI

A tree that can be embraced by two people grows from a tiny seed, and a nine-story tower starts from a pile of earth.

Halfway through 2021, the autonomous driving company Momenta , whose name is derived from the aforementioned Chinese classic , has been a hot topic in and outside the industry.

Because of its relationship with Great Wall Motors, some people call it the Chinese version of Cruise.

△ 5,000 units of Great Wall Motors' high-end "Mocha" vehicles have been equipped with the Momentum autonomous driving system

Because the first round of financing was an investment of several hundred million yuan by Shougang, Meituan and Hillhouse Capital, some people believe that it may be the fastest-growing driverless car unicorn.

With Zhang Kai, a Great Wall "veteran" with 20 years of experience, as chairman and Gu Weihao, the autonomous driving executive who knows cars best, joining as CEO, it is seen as a paradigm of planning before taking action.

But what is little known is that the slightest wisdom actually comes from adversity.

This is the worry of Wei Jianjun, the helmsman of Great Wall Motors, when sales volume rose against the trend last year and the profit margin per bicycle was the highest in the industry:

The dividends are about to end. With foreign brands in front and new forces behind, do we really have the killer weapon to win the future?

Momenta, with such a mission in mind, has opened up a new course as a "boat" for future transformation.

Because within a year, the "killer weapon" Momentum has delivered the following results:

The product has been mass-produced and installed in vehicles, with both soft and hard features.

In terms of revenue, it is expected to exceed 100 million within the year - such a growth rate sets a new industry record, and no autonomous driving company has achieved such a revenue level in such a short time since its establishment.

Haomo Zhixing Why? Let’s start with what it does.

What is Haomo Zhixing doing?

Gu Weihao , a veteran of China's autonomous driving and CEO of Momenta , admitted that before the financing news was announced, no one knew who Momenta was or what it was doing, and the recruitment presentation was also a bit "unfair".

△ Gu Weihao, CEO of Momenta

But at the beginning of 2021, Haomo Zhixing successfully became a hit with real money.

In addition to the major shareholder Great Wall Motor, investors such as Shougang Fund, Meituan, and Hillhouse Capital also participated in the financing of Haomo Zhixing , which completed the Pre-A round of financing with a financing amount of hundreds of millions of RMB .

Haomo Zhixing It has been less than 2 years since its establishment. What has it done to make the investment circle so optimistic?

In fact, to sum up, it is 134 .

1. A system that includes a full stack of self-developed autonomous driving software and hardware systems.

3. Three product lines, namely, passenger car assisted autonomous driving system , low-speed logistics delivery vehicles , and intelligent hardware , including service follow-up robots, cargo flatbed trucks, etc.

4. Four new report cards, corresponding to new products, new scale, new technology, and new cooperation.

First of all, there are new products, with a group of 10 autonomous driving products unveiled.

A typical example is the computing hardware platform - Little Magic Box 1.5 , which is used in passenger car autonomous driving. It will be mass-produced and launched this year and can support lidar solutions.

There are also the unmanned delivery vehicle "Little Magic Camel" and the unmanned vehicle's universal wire-controlled chassis "Little Magic Pan".

In addition, Haomo Zhixing also revealed that before the end of the year, Xiaomo Box 1.5 will be released. Xiaomo Box 1.5 is an autonomous driving product that specifically solves pain points such as congestion in urban commuting scenarios. It can achieve L3 assisted driving functions on highways and urban expressways, support drivers to take their hands off the wheel throughout the entire process, and add an industry-leading safety processing system. This product will gradually cover Great Wall's passenger car models.

Corresponding to the new product is the new scale .

In half a year, the mileage of the autonomous driving passenger car product "Little Magic Box" exceeded 450,000 kilometers, and 5,000 units of Great Wall's high-end model "Mocha" equipped with 100% of the Haomo Zhixing autonomous driving system have been installed.

It is expected that by the end of 2022, the Haomo Zhixing autonomous driving system will cover dozens of passenger car models of Great Wall Motors. Zhang Kai, chairman of Haomo Zhixing, disclosed that "it is expected that within three years, the number of passenger vehicles equipped with the Haomo Zhixing autonomous driving system will reach 1 million." By then, Haomo Zhixing will become the first company in the industry to reach this mass production scale in a short period of time.

In the field of unmanned delivery , Haomo Zhixing has truly realized the entire process of large-scale mass production and delivery. The number of Xiaomopan produced has exceeded 500. At present, Haomo Zhixing has built the world's first flexible manufacturing base specially built for L4-level intelligent service vehicles in Baoding, which produces a variety of models on mixed lines, covers an area of ​​5,000 square meters, and has a designed annual production capacity of 5,000 vehicles.

The new cooperation is to join hands with Meituan and Wumart.

Also within half a year, Haomo Zhixing has produced and delivered a large-scale complete low-speed logistics vehicle "Magic Bag 20" for Meituan, and completed the entire process of vehicle design - mass production and delivery, laying a solid foundation for the next step of large-scale mass production.

Wumart has installed Haomo 's autonomous driving system in multiple locations , and its logistics vehicles have now entered regular operation in Beijing's Shunyi area, serving nearly a thousand orders in total.

Of course, there are also advances in new technologies, which will be discussed in detail later.

But from these 134 , it should be clear what kind of company Haomo Zhixing is and what path it will take.

This is a route that technically captures the essence of Waymo + Tesla, with both platform-based high-dimensional technology and the advantages of mass production data iteration.

In business, this is a dual-wheel drive route of passenger transportation and logistics .

If we must find a comparison in the industry, combined with the background of the OEM Great Wall Motors, Momenta could not be more similar to Cruise, which is owned by General Motors.

In fact, as early as when Haomo Zhixing was officially launched by Great Wall and announced its independent financing and development, the Chinese version of Cruise had already been loudly called.

But the difference is that Momenta has stronger independent development capabilities and strength.

Because there is revenue.

How does Momenta do autonomous driving?

Dual-wheel drive, landing is king.

CEO Gu Weihao revealed: This year's revenue will be around 100 million yuan.

This is the confident answer given by the "newcomer" Haomo Zhixing after accurately sorting out the logic of autonomous driving.

This is also closely related to the fact that Gu Weihao is a veteran in the field of China's autonomous driving.

The first is the route selection for autonomous driving. Gu Weihao believes that the current technical routes of Robo Taxi and Robo Bus are not suitable for the survival of startups.

A number of domestic companies that started out in the Robo Taxi business in 2015 and 2016 have shifted their focus in the past two years. The fact that Waymo executives resigned one after another in the past two months, and that Lyft and Uber shut down their Robotaxi businesses, also proves that the progress of this sector is worth re-examining.

In fact, the reason is not complicated. Technology cannot guarantee that it will remain reliable after the safety officer is removed. If the safety officer cannot be removed, the cost cannot be reduced and the positive cycle of the business model is nowhere in sight.

The current algorithm is difficult to apply to manned vehicles traveling at a speed of 80 kilometers per hour.

But if we change our thinking, if we reduce its dimension and release it to a cargo platform with a speed of 10 kilometers per hour, its availability will increase a lot.

Therefore, low-speed logistics delivery vehicles have become one of Momenta's core businesses.

Moreover, compared with other players, Haomo Zhixing has obvious advantages.

We can develop a full set of self-developed software and hardware technologies, as well as the ability to produce and manufacture, the requirements have been verified, and the data can also serve the iteration of the system model.

CEO Gu Weihao's business logic is this: There are two types of customers with unmanned delivery needs. One is the Internet giants such as Meituan, Alibaba, and JD.com that have self-developed autonomous driving capabilities and are in urgent need of improving their hardware manufacturing capabilities. For them, Haomo Zhixing is a supplier of vehicles and wire-controlled chassis.

Another group of customers are supermarkets and express delivery companies such as Wumart, Hualian, and SF Express, which have delivery needs but lack the conditions for software development. For them, Haomo Zhixing can be a pure autonomous driving supplier or a comprehensive software and hardware solution provider.

This is also a key choice to unleash the organization's combat effectiveness, because with the unmanned logistics vehicle business, Haomo Zhixing has an independent business and business model, which comes from Great Wall but does not rely on Great Wall.

The most important thing is that there is now a consensus on the low-speed logistics vehicle track. If the cost of a single vehicle drops to around 100,000, the operating cost will be much more economical than the existing model.

As for passenger cars , Haomo Zhixing 's logic is still based on technical reality: Robo Taxi is difficult to implement, but it does not mean that the accumulated technology is worthless.

These technologies are not used to replace humans, but to enhance them, which is why Haomo Zhixing has always called its passenger car business "assisted autonomous driving."

The understanding of technology gives rise to the choice of business path, and the next step is "methodology".

Haomo Zhixing In summary, the first element of commercialization of autonomous driving is mass production .

Today, "three years of mass production" is becoming the recognized timetable for the implementation of industrial technology in the autonomous driving industry, but very few companies can actually achieve this.

Haomo Zhixing has an advantage that other players cannot match: it has the Great Wall behind it .

As mentioned before, almost all Great Wall models will be equipped with the Momentum Intelligent Driving autonomous driving system in the future, which will solve the problem of increasing the volume of autonomous driving systems in the short term and have the same advantages as Tesla.

In addition, this model also breaks through the shortcomings of Waymo's road test areas and limited mileage, which can only be compensated by time.

With mass production, massive amounts of data will continue to flow in, and Haomo Zhixing will be able to quickly establish its technological iteration advantage over other competitors.

As for data, Haomo Zhixing also has a different set of "usages".

Data capabilities determine success or failure, and "great effort brings about miracles"

Regarding the "second wave" of autonomous driving craze that began last year, Gu Weihao saw the following trend: autonomous driving sensor solutions and computing platforms are becoming increasingly similar, and there are only a few suppliers, which cannot widen the technological gap.

Therefore, in the future, the key to success for autonomous driving companies will be data . Whoever has more data and lower data acquisition costs will have the best chance of outperforming its competitors.

This is also the new technology that Haomo Zhixing disclosed this time.

Haomo Zhixing Self-developed autonomous driving system, passenger cars and logistics vehicles, in this system, share data collection, generalization training and deployment.

Simply put, there was a library of one billion samples at the beginning of training, but we found that some problems were not solved, such as stopping when encountering a water bottle on the road.

The water bottle is not in the training database, so we need to query and search the image of the water bottle and all the data collected in history, and add all the data related to the water bottle to the database.

Next, we will find more pictures of water bottles in more databases, perhaps tens of billions of them, and add them to the training sample library.

The next step is to balance the data samples, perhaps by adjusting some of the thresholds or transition algorithms, or small operators, and retraining.

Therefore, at the data level, passenger cars and logistics vehicles are shared, and the only difference is the adjustment of the dedicated algorithm model.

With high-quality data acquisition and application, Momentum’s focus on autonomous driving has evolved from “perception” to cognition .

Human drivers can basically judge at what kind of intersections pedestrians may suddenly jump out, which non-motor vehicles may violate the rules, etc. This ability is based on human high-level cognition.

Through prediction, we will focus on directions and distances where problems are likely to occur to ensure safe passage.

Similarly, the same principle applies to the creation of AI drivers.

The purpose of machine learning is to learn an optimal model from the training data set, and this optimal model has good predictive ability for unknown data (that is, the model has good generalization ability), thereby achieving a breakthrough from perception to cognition.

Cognitive intelligence has very high requirements for data scale, acquisition cost, data quality, and system data processing capabilities. Haomo Zhixing ’s self-developed system precisely meets these requirements.

As for "great effort brings about miracles", this is a cutting-edge technology that Haomo Zhixing is currently tracking, which was introduced by Gu Weihao : Transformer .

Gu Weihao himself has a background in technology research and development, and as a CEO today, he remains a hard-core person. In particular, he has unique and sharp insights into the development history and prospects of computer vision technology.

△ Gu Weihao, CEO of Momenta

Readers who follow the frontiers of machine learning are certainly familiar with it. Transformer was first designed for language processing tasks. It has a model structure that avoids recurrent cycles and completely relies on the attention mechanism to model the global dependency between input and output.

In other words, as long as your data is large enough, you can train a super large model.

In the past two years, Transformer's accuracy, efficiency, and robustness in image recognition have continuously broken various charts, leaving the rest far behind.

This is a true miracle of great effort , because this method requires a huge data set, which perfectly matches Haomo Zhixing 's advantages in data collection and generalization.

Finally, let me summarize the logic of Haomo Zhixing ’s autonomous driving startup.

The core idea is to seize the initiative in the implementation of L4 autonomous driving scenarios and scale, while also laying out three mass production lines for unmanned delivery vehicles, smart hardware, and passenger car autonomous driving.

Then, the autonomous driving data in real scenarios is continuously supplied to train a high-level autonomous driving system with universality and scenario generalization capabilities.

Ultimately, through progressive function and scenario coverage, large-scale commercialization in multiple scenarios will be achieved.

Self-driving car startups have a new paradigm in China

Gu Weihao’s sharing on recruiting is very interesting.

Previously, few people were interested in the recruitment fairs, but today, many outstanding technical talents in the industry have taken the initiative to apply, including some from large companies.

Why choose Haomo Zhixing ?

Gu Weihao gave the most direct and realistic reason: technical talents believe that Haomo Zhixing ’s option incentives are feasible .

In other words, according to some of the technical R&D personnel who have joined Haomo Zhixing , Haomo Zhixing is not "painting the cake".

This is an acknowledgement of Haomo Zhixing’s business path, and this acknowledgement is also based on the perspective of technological reality.

Of course, there are also some people who join Haomo Zhixing because they recognize the team and the technology itself. For them, recognizing that the company has a good team and is making a good product is enough to impress them.

△ Momenta’s core executive team

Therefore, the potential and value of Haomo Zhixing have at least the first level of meaning, which is to define a new "success model" for autonomous driving startups.

On the surface, this model seems to be a promising business - the OEM is the major shareholder and customer, so there is no need to worry about funding.

But the key is to open up a scenario that can be quickly implemented and scaled up, and then design your own efficient data solution based on the characteristics of the scenario, to avoid years of testing in a limited venue and delaying implementation.

Haomo Zhixing Chairman Zhang Kai gave an "official explanation": Haomo Zhixing is the Chinese representative of the "car company + technology company" Cruise model.

Moreover, Haomo Zhixing is taking a gradual commercialization route, with AI "cognitive intelligence technology" as the core of its research and development.

△ Zhang Kai, Chairman of Momenta

Haomo Zhixing was born from the combination of the traditional automobile industry and autonomous driving algorithms and software capabilities, so the second meaning may be that Haomo Zhixing has the potential to become the first "showroom" for the transformation of traditional car companies.

Last year, the then Volkswagen CEO Diess admitted that Volkswagen's self-developed software rate was less than 10%, and ICT technology was an almost insurmountable gap for traditional car companies.

Not only is there a shortage of programmers, but there is also a shortage of executives who understand AI research and development. Even if an excellent technical team is recruited, it will be difficult to manage it well. Moreover, the incentive mechanism of traditional car companies makes it difficult to attract such talents.

Haomo Zhixing is another attempt.

Although Great Wall Motors is the major shareholder, Haomo Zhixing has maximized its enthusiasm in organizational design and is placed outside the "greenhouse" of the entire listed group, making it relatively independent.

Great Wall hopes to use the efficient management and incentive methods of an Internet technology company to cultivate an intelligent killer weapon that can operate independently and achieve an independent listing.

Haomo Zhixing While "giving back" to Great Wall, it is also an important player in the autonomous driving track, exporting technical solutions to any potential customers.

In the dual-wheel drive of Haomo Zhixing , the first customer of mass-produced passenger cars is Great Wall Motors, but Haomo Zhixing firmly believes that with its technological strength, more car manufacturers will come, and it is a technology supplier.

In the mass production of unmanned delivery business, Haomo Zhixing can also provide end-to-end software and hardware integrated solutions and can be a product company.

In terms of this business model and company form, there is indeed only Cuise under General Motors in the world, but Cruise is more oriented towards passenger and manned vehicles.

However, for Cruise, GM first took it back and then "released" it, while Haomo Zhixing was planning for commercial implementation from the beginning and promoting three product lines at the same time...

The awareness of "worry" is also inherited and emphasized in Haomo Zhixing . Gu Weihao said that Haomo Zhixing has been "running for its life" in order to survive.

Therefore, the "startup" company Haomo Zhixing has produced 10 different product categories in two years, and the mass production progress is also advancing rapidly.

The so-called Cruise autonomous driving entrepreneurial model, which is different from Waymo and Tesla, may be interpreted into a more advanced and mature version in China.

And now, the fledgling Haomo Zhixing has already shown its characteristics:

Momenta is China's fastest growing driverless car company.