Why is autonomous driving bottlenecked, but automated valet parking is hot? ｜Miss A Lightyear

Writing | Match Q

Editing | Ms. A

Design | Yifan

In the video, a woman driving an Audi A8L gets out of her car near an office building, and she gently presses a key on her phone, and the car comes to the basement and enters a parking space on its own without anyone driving it.

When the woman left the building, she simply used a key on her phone to summon the car at the pickup point, and the car drove back from the garage on its own.

Open the door, get in, and go, cleanly and efficiently, eliminating the time-consuming and labor-intensive process of parking and picking up the car.

This is the "Auto Valet Parking" (Auto Valet Parking) scenario (in Audi's display at the time, the concept of piloted parking was used).

Functionally, it is an upgrade of Park Assist, ADAS (Advanced Driver Assistance System), and so on; technically, it belongs to the low-speed L4 level of autonomous driving.

But five years on, the not-so-new concept of AVP has yet to shine into reality. In the past two years, the automated driving boom is more manifested in another segmented track: L4-level automated driving for passenger cars on urban roads.

This is the most in line with the general public's impression of automated driving - without the need for a driver to operate, the car can automatically carry passengers through the streets and alleys, and complete the task of overtaking, reversing, avoiding pedestrians, and other tasks that should be accomplished by a human driver.

The ultimate business model is Robo-Taxi, which is part of the TaaS (Transportation as a Service) model. Founded in 2009, Waymo (part of Google's parent company Alphabet) is a pioneer in this space.

From 2016, this sub-track is getting hotter and hotter: General Motors acquired Cruise, an autonomous driving company, at a valuation of $14.6 billion; Drive.ai, Pony.ai, Rodarstar.ai, and Wenyuan Zhixing have set up startups aiming at high-speed L4/L5-level autonomous driving; Baidu's Apollo Baidu's Apollo autonomous driving platform, Uber and Dripping's autonomous driving program has been launched one after another.

But this "rapid development" seems to be mainly in financing and valuation. After the boom of the first two years, the L4 level autonomous driving of passenger cars on urban roads entered a bottleneck in 2018.

In November 2018, John Krafcik, CEO of Waymo, confessed in public that the popularization of self-driving technology will take decades. A few days later, Apple co-founder Steve Wozniak also said that self-driving cars are unlikely to be realized in the near future .

These world-class minds are stuck on two words: mass production and commercialization.

AVP, however, is likely to be the first to break the deadlock on mass production and commercialization of high-level autonomous driving for passenger cars in urban spaces.

Now, the market is finally just a step away from the mass production of AVP.

One of the companies that may be able to make this "kick in the pants" is Zongmu Technology, a Shanghai-based company that has entered the field of automated driving from ADAS assisted driving systems.

On December 26, 2018, Zongmu Technology announced that it had received a project designation letter from China FAW, and would deploy low-speed L4-level autonomous driving AVP products in the 2020 mass-production models of FAW's Hongqi brand. Longmind is the first company in China to win an OEM AVP mass production project.

What made Zongmu take the order from the OEM is the upgraded version of AVP version 1.0, which they launched in November last year for low-speed scenarios in parking lots.

At that time, AVP was not the center of attention in the domestic autonomous driving field, compared to the direction of unmanned vehicles, which has a stronger sense of black technology.

It was only in July 2017, when Daimler and Bosch announced the concept of Auto Valet Parking at the Mercedes-Benz Museum*** in Stuttgart (their collaboration in this area began in 2015), that the industry had a more unified view of the scenario.

Auto Valet Parking

The concept of Auto Valet Parking (a partnership between the two started in 2015) has led to a more unified approach to this scenario.

And in 2018, the AVP competition is already in full swing.

Looking at the above progress, most of them are the release of concepts, demos, technical solutions, and strategic cooperation.

This suggests two things:

First, each player sees the prospect of AVP.

Tang Rui, the founder and CEO of Zongmu Technology, also told Azi Lightyear, "I think AVP is the most strategically valuable high ground in the Chinese market [in the field of autonomous driving]."

Prof. Dr. Hans Georg Engel, head of Mercedes-Benz Passenger Vehicle R&D Center in China, also said last year that AVP is an important milestone on the road to autonomous driving.

Two, there aren't many players who really have a long history in this space.

Some of the above developments were initiated only last year, and many of them are only at the stage of collaborative releases, not yet at the testing, much less the mass production stage.

Against this backdrop, the AVP mass production order for FAW Hongqi sedans that Longview received in December is of great significance in terms of market development -- the AVP system of Longview is likely to be the first one to be registered in mainstream passenger cars.

At a time when many players are piling into AVP, Tang Rui, the founder and CEO of Admiral Technology, reviewed his thinking to AZLN: Why did he look at AVP in the first half of 2017?

We can look at this question from the perspective of business logic first.

When considering the size of the opportunity and the commercialization speed of each landing scenario of autonomous driving, we mainly look at four dimensions: closed scenarios/open scenarios, fixed routes/free routes, low speeds/high speeds, and passengers in the car/no passengers.

In addition to the technical differences, the above four dimensions, closed scenarios/open scenarios also involves the right of way; with passengers/without passengers involves the difficulty of legislation.

There are two main business models for autonomous driving:

One is the TaaS mentioned above, in which autonomous driving is the core technology that enables the establishment of operational services.

The second is the traditional business model of the automotive industry, which is to add self-driving capabilities to a car and sell it to consumers, which Tang Rui summarized as VaaP (Vehicle as a Product). In this model, the company providing the autonomous driving system acts as a supplier to the OEM.

VaaP is the present, TaaS is the future.

With this framework, the most complex and commercially valuable scenario is "open roads, free routes, high speeds, and people in cars", which can be applied directly to urban mobility - Roadstar.com. Roadstar.ai, Pony.ai, Wenyuan Zhixing, and Waymo, the world's top self-driving company, are all challenging this "laurel scenario" . The corresponding business model, Robo-Taxi, will revolutionize the way people travel, and then change the way they live and interact, giving rise to a series of new businesses.

This is also the reason why these companies have attracted a lot of venture capital, and the big opportunity is not to be missed. But the time for commercialization of this scenario seems far away at the moment.

The easiest scenario is "closed roads, fixed routes, low speeds, and no one in the car", such as mining trucks in mining zones and unmanned logistics vehicles in parks; however, the commercial value of this scenario is limited, and the migratory nature of the scenario to other scenarios is limited.

AVPs, on the other hand, are in the middle - "open road (but not many pedestrians in parking lots, especially in types of parking garages such as basements), free routes, low speeds, and no one in the car".

This is just a compromise scenario with a lot of extensibility, and it has the following characteristics:

I. Small legal and regulatory barriers - fast landing, high volume

Compared to other urban

In this scenario, there is no ethical dilemma of whether to protect passengers or pedestrians, and the risk of accidents is lower due to low speeds and fewer pedestrians in and out of basements.

Tang Rui told A Zi Guang Nian that he thinks the fundamental reason AVP can be implemented faster is that "the laws and regulations are simpler, and the willingness of car makers to place mass production orders will be stronger."

The direct business consequence of this is that AVP can land faster and in large quantities.

Second, for both the VaaP and TaaS markets - able to make blood, large space

In the VaaP market , Volkswagen, Daimler, BMW, Volvo and other companies have embedded AVP systems embedded in the production models on the agenda. The FAW AVP mass production order that Zongmu got this time also belongs to the VaaP model.

VaaP allows AVP manufacturers to generate revenue in the relatively short term, and has a certain degree of self-supporting capability.

And at the same time, AVP has the potential to cut into the operational services, or TaaS, market.

Now, the AVP product form factor can go directly to the car timeshare operation service.

This model is actually the car version of the bicycle. But compared to a bicycle, a car is more complicated to pick up and drop off. Time-share operators need to spend a lot of operating costs on scheduling vehicles, and the inconvenience of parking and picking up the car also affects the market acceptance of time-share leasing.

So there is a strong demand for time-share operators to adopt AVP systems.

In the time-share model, AVP vendors have the opportunity to participate in the operation and provide value-added services, such as providing data insights for the insurance industry, adding precision marketing, destination discount information, advertising, and other businesses to the passenger's AVP-controlled portal (e.g., mobile app).

This is a business model with a lighter role, greater marginal benefits, and more profit margins than that of a supplier.

Third, AVP has the potential to cut into the "laurel scene" in the future

As mentioned earlier, the capital and talent invested in autonomous driving is huge, and the most exciting thing for everyone is the "laurel scene".

As mentioned earlier, there is a lot of capital and talent invested in autonomous driving, but it's the "laurel scenario" that has everyone most excited - the high-speed autonomous driving scenario on city roads.

AVP has the potential to cut into this scenario.

That's because AVP is actually very technically demanding.

This scenario meets the characteristics of open roads and free routes, which means that full L4-level autonomous driving technology is required to ensure the large-scale commercial landing of AVP.

The differences between AVP and autonomous driving laurel scenarios are mainly:

And the above two major differences are the bottlenecks that currently limit the landing of high-speed L4-level autonomous driving.

The former makes it necessary to use LiDAR in the program, which has a longer detection range but is expensive and costly.

The latter is a technology breakthrough. As Waymo CEO John Krafcik said, it will take decades for autonomous driving to become commonplace, mainly because the technology has not yet broken through to the highest level of L5, which allows driving in all weather and under all conditions. as long as this ability is not unlocked, the safety of autonomous driving cannot be truly guaranteed.

In other words, companies with the strength to make a reliable AVP program are not technologically separated from those aiming for high-speed L4-level autonomous driving.

In Tang Rui's words, the technology is there. But the technology that should be there is not there yet, we are still waiting.

Secondly, compared to the segmented tracks such as park logistics vehicles, highway trucks, etc., AVP is directly applied to passenger cars, which is more commercially and cooperatively compatible with high-speed automated driving scenarios for passenger cars.

Becoming a low-speed L4-level AVP supplier for OEMs first, and then **** the same R&D for high-speed L4-level autonomous driving, could be a steady path to the laurel scene.

"I think this is a strategic high ground, if this occupied, and then go to fight high-speed, this is very Solid (with a solid foundation)." Tang Rui said, "In China, low-speed L4-level autonomous driving may land first."

The fact that Longitudinal Eye was able to capture the AVP opportunity earlier in late 2016 is also due to the genetics of this company.

Longitudinal just represents one of the major schools of thought in the field of automated driving, "car + AI", and the other major school of thought is "AI + car".

The so-called "car + AI" is to use the automotive industry as a base to cut into autonomous driving. Attempts by car manufacturers such as GM, Volkswagen, and Toyota, and Tier 1 (Tier 1 suppliers) such as Bosch, ZF, Continental, and Hyundai Mobis belong to this school of thought, which is characterized by product orientation.

Founded in 2013, Vertical Vision is also a "car school" autonomous driving company, with a deep automotive background:

Tang Rui, founder of Vertical Vision, has 14 years of automotive semiconductor R&D and management experience; the team also includes several international executives and senior executives from Tier 1, such as Chen Zhuochao, chief engineer of Chery Automobile Engineering Research Institute, and many others. The team also includes a number of international Tier 1 executives and technical experts such as Chen Zhuochao, former Chief Engineer of Chery Automobile Engineering Research Institute, and other AI talents such as Wang Fan, who has worked as a senior R&D manager at Zoran and CSR, after starting to set up the technical team in 2016. Zoran has a clear and lucid understanding of the mass production process of car manufacturers and how to control product costs.

AI+Vehicle" is based on AI technology, starting from the algorithm of automatic driving, and gradually seeking commercialization. Many companies with an Internet technology background belong to the "AI+Car" school.

Overseas representatives are Waymo, Cruise, Uber Autopilot, and Auto X. Domestically, Baidu Apollo and Pony.ai, Roadstar.ai, and Wenyuan Zhixing (the last three companies mentioned have a Baidu background).

But by the time we get to the depressed environment of 2018, the market is more concerned about who can live well in the present.

In terms of "survivability", the car + AI faction has a certain advantage.

Throughout the history of the development of this is also the case.

2013 to 2017 is the first stage of the vertical program, completing the leap from startups to Tier 1.

At this stage, the main products promoted by Admiral are L0 to L2 ADAS systems, which can realize the driver's assisted parking and assisted driving functions in the last 3-5 meters.

By May 2017, Zongmu, which has established its own production lines in Xiamen and Shanghai and received IATF16949 certification (a production process quality system certification), has upgraded from a Tier 2 supplier that mainly provides algorithms and solutions to a Tier 1 supplier that is directly oriented to car manufacturers and provides software and hardware products and services at the same time, and cooperates with car manufacturers including Geely, Weimar, SAIC Datsun, The company's partners include Geely, Wima, SAIC Datsun, BAIC Yinxiang, Jiangling New Energy, and Chehejia, among others.

At the same time, after the completion of the A round of financing, Zongmu began to set up an automatic driving team after the Spring Festival of 2016. The biggest difference with the "AI+Car" faction is that from the first day of autonomous driving, Zongmu thought about how to really mass produce.

Tang Rui has three thoughts on how to realize the mass production of autonomous driving.

First of all, mass production has to be "automotive grade" in terms of quality.

A car is a means of transportation, but from another perspective, it is also a "killing machine". The reason why cars have become the jewel of the industry is that any parts and systems used in a car must be subjected to a variety of extreme conditions such as high and low temperatures, drops, vibrations, and aging, to ensure safety and stability as much as possible.

It's a process that can't happen fast enough. Gu Jianmin, CTO of Valeo China, once said, "From the program to the final mass production, the development of a real automotive grade usually takes five or seven years." As Tier 1, when Zongmu started to do automated driving, it had already shipped 500,000 sets of automotive-grade products, and had rich experience in R&D and mass production.

Secondly, while quality is excellent, costs must be strictly controlled in order to get to volume.

This is also the reason why Zongmu is focusing on low-speed L4 scenarios, because low-speed scenarios don't require long braking distances, don't need to use tens of thousands of dollars of LIDAR and other expensive parts, and don't have high requirements for the processing power of the computing platform.

Vertical vision based on the perception of the vision of the solution, with low-cost millimeter wave radar, the use of lower-priced Qualcomm Snapdragon 820A platform, the cost of AVP to control very low. "We can get the whole system, controller plus all the sensors to within a few thousand dollars." Chaozhuo Chen, vice president of Vertical Vision, told "AZ Lightyear".

Third, mass production also needs to meet market demand.

Difficulty in parking, difficulty in picking up a car, and congestion on urban roads are special pain points in China's travel sector. So Tang Rui judgment, the United States may have a certain advantage in the high-speed scene, but in China, the probability is that the low-speed L4 autopilot first landing.

Under such thinking, Admiral began to develop AVP products in September 2016, and at the end of November 2017 took the lead in launching its own AVP1.0 system.

Looking back on how Zongmu was able to gain a head start in the market, Tang Rui said, "It's a matter of mindset, At that time, a lot of people wanted to be very cool high-speed automated driving companies, and didn't seriously think about what the Chinese market actually needs for automated driving. "

In the specific opening of the AVP business approach, Longview also maintains the "car + AI" steady .

Steady is first shown in the capital borrowing. Because the ADAS product line is relatively mature, unlike many autonomous driving companies that raise large amounts of money, Longitudinal's approach is to cover part of the cost of R&D for autonomous driving with the blood production capacity of ADAS, and therefore does not rely as much on external capital.

In terms of expanding the business model, Zongmu is still doing a good job as a supplier, and has been promoting AVP mass production orders with OEMs since early 2018, after technical exchanges, SOR (Specification Of Requirements), tender, finalization, and the finalization of bidding, Specification Of Requirements (Specification Of Requirements) for suppliers. , bidding, bidding, factory audit of the complex process , and finally at the end of the year to get the FAW's fixed point letter.

During this process, the advantages of the Tier 1 supplier role that Zongme spent more than 3 years to establish began to emerge, which is specifically manifested in the good business relationship with the purchasing department of the automobile factory, the technical cooperation with the automobile factory research institute for many years, and the tacit understanding with the SQE (Supplier Quality Engineer) in the production and quality control.

At the same time, the automatic driving technology also allows Zongmu to jump out of the "first-tier suppliers" to realize the industrial chain position of the opportunity to upgrade, that is, to cut into the time-share rental operation services.

"I have a point of view, the L4 level of automatic driving must be operation-related. Tang Rui said, "Timeshare is a big opportunity we see, it not only upgrades the traditional ADAS, but also brings new business models. "

At present, Zongmu has already made initial contact with the time-share platforms of several OEMs, **** with exploring the way AVPs can be landed under the time-share model.

Combing through the 5-year history, the not-so-high-profile Zongmu has a clear upgrade path from assisted driving to automatic driving:

"From the whole company, the strategy is very clear - how do we set up Zongmu's leading position in automatic driving from the low-speed, AVP this cut. position. This we have not changed." Tang Rui told AZLN.

On the eve of AVP mass production, there are still some uncertainties in the market.

The biggest point of contention is the direction of convergence of technical solutions:

Currently, there is a spectrum of distribution on what to do with AVPs, with the "car conversion school" at one end of the spectrum and the "garage conversion school" at the other.

At one end of the spectrum is the "car conversion school" and at the other end is the "garage conversion school".

The vertical eye is the "car conversion school", which pursues "car intelligence", i.e., realizing AVP without converting the basement.

Tang Rui believes that there are three benefits to converting the car solution:

First, it can better control the cost. --The cost of converting a garage is relatively high, and it limits the geographical scope of AVP use;

Secondly, Zongmu itself has a strong partnership with car factories, while large car factories can quickly catalyze the commercial feature of AVP through mass production and shipment, and converting a garage involves cooperation with real estate developers and properties, and the market fragmentation and

"If the parking lot has communication and car - field synergy is better, but if not transformed, the car also has enough intelligence, just like people can find parking spaces in strange parking lots." Tang Rui said.

Horizon also belongs to the modified car faction, on November 28 last year, Horizon for the first time publicly tested its AVP project, but Horizon's current program uses Velodyne 16-line LiDAR (officially priced at $3,999), and the cost is yet to be further reduced.

Bosch, the world's largest tier-1 supplier, is a representative of the "change garage school" , through the transformation of the garage, so that different models to realize the "auto-parking", the vehicle itself only needs to be electronic. The vehicle itself only needs electronic braking, automatic gear shifting, electronic power steering and remote connectivity and other basic functions, without any environmental awareness functions.

The advantage of this solution is that it is less demanding on the vehicle, but the cost is currently higher.

In the joint Bosch and Daimler demonstration in China last year, they used a column-mounted, industrial-grade, single-line LiDAR. Based on system redundancy considerations, the actual application, may be an average of every 3 parking spaces to layout 25 LiDAR, in the cost is not yet realistic.

The Holoparking that WoD launched at the end of last year is in the middle.

Summarized in the words of Wodo's founder, Kai Ni, it's a "three-in-one" solution consisting of "field end, high-precision mapping end, and vehicle end".

The difficulty of this program is the need to do systematic research and development: "I need to do research and development on all three ends, the surface is relatively open, and requires a lot of technical modules **** the same to complete." Ni Kai has said in an interview.

Regardless of whether it is "change car faction" or "change basement faction", in the process of promoting the commercial landing of AVP, are faced with a **** the same problem, that is, the need to integrate the resources of all parties, to develop a set of new industrial chain benefit The reason for this is that AVP as a L4 driver is a very important part of the L4 system. This is because AVP, as an application of L4 automated driving, has broken away from the concept of a single hardware module or software function; the AVP technology provider is no longer just a Tier 1 supplier for OEMs, but is more and more y involved in the whole life cycle of AVP "product + service".

Currently, there are six types of commercial players involved in this AVP scenario: AVP automatic driving technology providers, OEMs, time-share operators, high-precision map makers (the realization of automatic driving technology requires supporting high-precision maps in the basement), charging technology providers, and commercial real estate/property companies.

The issues that remain to be clarified by all parties are mainly twofold.

One is the business model, which involves who ultimately pays the bill and how the upstream is split.

The buyer is relatively clear - whether it is VaaP mode or TaaS mode, the final payer is the driver/passenger at the C-suite; and under the TaaS mode, there can be part of the "wool on the pig" revenue. "Revenue, such as marketing, advertising, insurance and other businesses combined with travel services.

What the players need to consider is how much the consumer at the end of the line is willing to spend on an AVP system? Are they willing to spend money on a per trip basis, or are they paying all at once?

The upstream profit-sharing model has even more uncertainty. For technology vendors like Zongmu, they may want to end the old single charge model of selling products and licenses, and instead charge per call, and try to cut into AVP-based value-added services; map vendors and properties will also need to find a reasonable way to share profits in the new model.

The second is how the rights and responsibilities are distributed: In case of an accident during autonomous parking, who is the responsible party? How should the insurance company pay out?

"It must be a business model of service," Tang Rui firmly optimistic about the combination of AVP and time-share operation services, but he also admitted: "The specific distribution of benefits, the distribution of responsibility is not clear to see now."

But this "fuzzy" state often means opportunity.

For latecomers to the automotive industry like Admiral, they have the opportunity to gain the most favorable market position through the new round of changes.

"The whole market is relatively fragmented, no one has a complete say, and it is because this is a complex matter that a core technology provider needs to take the lead. "Tang Rui this view of the current situation and the possible location of the vertical eye.

Currently, Vertical Vision has been discussing new business rules with some of the car makers, figure dealers, commercial real estate properties, and other relevant players.

Next, the long term goal of Admiral is to upgrade to high speed L4 level autonomous driving based on the realization of AVP.

The step-by-step company has developed a three-step program.

The first step is to continue to work closely with OEMs to run through the AVP technology mass production landing from 0 to 1.

In 2018, Zongmu has received a large number of pre-research projects from domestic car companies, and after the mass production AVP order for FAW Hongqi models, Zongmu may announce more good news, including entering the supply chain of joint venture car companies.

Tang Rui's own judgment is that the key to running through from 0 to 1 is 2020, when the vehicle loaded with AVP reaches more than 100,000 units a year, AVP is considered to have withstood the market test.

At the same time, Longview has also begun to layout the second step, using AVP technology to improve the scheduling efficiency of time-share service providers.

"I personally judge that companies doing time-share rental are more suitable for copying from 1 to N." Tang Rui told "A Zi Guangnian", this kind of operation manufacturers are more important to market promotion and volume, before the product is fully verified, it is difficult to meet their commercial expansion needs.

Therefore, the plan of the vertical program is to start talking about specific landing projects with time-sharing companies at a slightly later time - probably in the second half of 2019.

Before that, Admiral will run through the various technical points of combining AVP and time-share leasing through a large number of tests on a small scale, and then go to do joint operations with large platforms.

The third step is to begin to enter the field of high-speed L4 autonomous driving in the future.

In this process, Zongmu will still leverage its Tier 1 status to gain revenue from its ADAS and AVP product lines, which will be self-supporting.

At the same time, when the technical conditions are more mature, the cost of sensors is reduced, and the legal and regulatory environment is more perfect, it will enter the "laurel scene" of high-speed L4 autonomous driving.

Tang Rui's judgment on the "timing" is that the competition for real L4-level products will occur in the next 5 to 10 years, which is a big enough opportunity, but also need to pay enough patience.

"The thing that really makes autopilot exciting must be service." Tang Rui said.

In the new era of combining autonomous driving and operations, Admiral will not stay in a traditional Tier 1 role, but consider more possibilities - such as cutting through core components to y participate in a series of value-added services brought about by changes in the mode of travel.

"We want to be the first company in China to prove that L4 can really be commercialized and landed." Tang Rui is a person who does not want to show too much front, but this time, he did not mince words that the vertical eye is to be the first.

Who will be the first, like the development of autonomous driving, "speed is not enough", and you can't see the results in a hurry.

But being the first to get an order from FAW for AVP mass production has added credibility to Indulge's boast that the company has won a favorable position. In the next two years, it will launch a sprint to the stage of victory.