Autonomous driving in the eyes of the public is being cloaked in danger by deadly marketing.
Tesla Model 3 electric car crashed into the truck and killed the owner; Waymo unmanned car just melted 2.5 billion and hit people in San Francisco; 31-year-old entrepreneur Kai Weilai self-driving car accident died...
Although big data proves that autonomous driving is safer than human driving, the frequent occurrence of car accidents is inevitably daunting.
Focusing on the competition in the second half of autonomous driving, there are not only safety dilemmas, but also large-scale mass production, commercial landing and other obstacles.
As Mentioned by Xin Guobin, Vice Minister of Industry and Information Technology at the 2022 China Electric Vehicle 100 Forum, in terms of supply chain and industrial chain, it is also necessary to deepen integration and innovation, strengthen the innovation and construction of manufacturing centers, accelerate the research and industrialization of key technologies such as new system batteries, vehicle specification chips, and vehicle operating systems, promote the integrated development of vehicle and road network cloud maps, accelerate the construction of new information network road infrastructure, and expand the multi-scenario demonstration application of intelligent networked vehicles.
There is no doubt about the huge profit margin of unmanned driving, provided that the premise is how to smoothly survive the pre-dawn darkness.
The combing around the robotaxi landing route, the autonomous driving technology route and the perception device may help us find a commercial road for intelligent driving based on the premise of safety.
Giant Finance "Industry Observation" Part 22.
Lead Writer / Abscisic Acid
Article Architect / Quiet
Producer / Giant Finance
01
Robotaxi is behind the close-quarters hand-to-hand combat
Technical feasibility does not equal commercial closed-loop
The current Robotaxi is still the focus of firepower in the autonomous driving subdivision track.
Big factory travel party
Baidu can be called the "leading big brother". Up to now, Baidu Apollo's self-driving travel service platform "Radish Run" has been launched in 8 cities, and Q4 has provided 210,000 ride experiences.
Didi takes the research and development of autonomous driving as the highlight and constantly completes the hardware capabilities. Obviously, for this team of about 500 people, it is not just money that needs to be added.
"Companies that do Robotaxi are doomed" and "Tesla self-driving kills people." The commercialization of the manned Robotaxi has been delayed, but neither the dangerous speech of former intelligent driving minister Su Zhen nor the valuation of technology pioneer Waymo can continue to spend money on Chinese players.
Traditional car companies are heavily positioned. SAIC Motor launched the first domestic car company L4 autonomous driving operation platform Xiangdao Robotaxi, and the Robotaxi model jointly developed by Beiqi Blue Valley and Baidu was mass-produced.
Robotaxi shorts of engagement. Today's competition on the field not only has traditional car companies and big manufacturers travel school become the forerunners, but also the head entrepreneurial school to catch up.
Head entrepreneurial
Chima Chi Heng, who has just completed the first closing of the Series D financing, plans to expand the team, increase technology research and development, and expand the scale of robotaxi and Robotruck fleets.
The IPO was stranded, and the backbone of the department successively left the Xiaoma Zhixing business at the same time, while Wenyuan Zhixing, which explored the formation of an iron triangle model with car companies and platforms, is cooperating with GAC Group in depth.
After the strategic investment matters were finalized, the two sides worked with Ruqi Travel to start the design and development of Robotaxi front-loading models and upgrade the mass production landing.
The Robotaxi landing route of L4 before 2019 is fragrant. Represented by Waymo, baidu, Xiaoma Zhixing, and Wenyuan Zhixing in China basically take this route, from research and development, road test operation to scale landing to directly climb the highest peak of L4.
Capital investment in the L2 sector is slightly deserted, but the competition is equally fierce. Represented by Tesla, Wei Xiaoli is inclined to mass production and iterative mode after the road, that is, from L2 to L4 step by step.
New energy vehicle companies
Judging from the research and development plans of the three new car-making forces, they are trying to create a full-stack self-developed autonomous driving capability.
The difference is that WEILAI is trying to build a high-end brand with community and service, ideally inserting a horizon "Journey 3" platform in the process of switching from Mobileye to NVIDIA, while Xiaopeng, who has always been focusing on automotive technology, has also accelerated its speed towards the Robotaxi track.
Robotaxi is one of the best ways to accumulate test miles and improve the stability and security of your algorithms. However, due to the commercialization process, the exploration of autonomous driving is actually tantamount to experiencing a marathon long-distance run for all players.
The first half of autonomous driving verified the feasibility of the technology. "From exposing one technical problem per few kilometers traveled in the past to one or more critical problems that may require traveling tens of thousands of kilometers in the future."
Although it sounds like the problems of the technical route have decreased, there are still two major roadblocks in front of Robotaxi.
One is the robotaxi profitability issue, which is obstructive and long. After all, even the automatic driving of freight that is easy to mass-produce, such as the operation efficiency and cost of unmanned delivery vehicles, cannot reach the level of manual distribution.
The other is about data collection. All the competition for autonomous driving, in the final analysis, revolves around effective data collection and driving AI model iterations. The construction of urban roads in China is very fast, and it is not easy to achieve data collection and update of high-precision maps.
02
Baidu's fans of the road to car coordination
More in line with national conditions
Optimistically, with the improvement of technology and policy opening up, China's autonomous driving has ushered in a new development node.
Moreover, combined with the development status of the industry's three major technical schools (bicycle intelligence, vehicle-road collaboration, bicycle intelligence + vehicle-road collaboration) and national conditions, the full-stack service route may become the key to grabbing the second half of the ticket.
Enterprises with a focus on bicycle intelligence: rely heavily on high-performance AI chips and algorithm support.
In the sparsely populated United States, where road conditions are less complicated, it is not surprising that private companies responsible for building communication networks bet on bicycle intelligence out of input-output considerations. Companies such as Tesla and Google choose to take this route based on the advantages of chip and algorithm technology.
In contrast, in China, although there are enterprises such as Wenyuan Zhixing and Tucson Future that emphasize the technology of the car, due to the demographic road conditions, the difference between the construction of 5G networks, and the weaker chip and software development capabilities, many companies have turned to network empowerment.
Focus on network-enabled enterprises: it can be subdivided into two categories.
Category 1: Enterprises such as Datang Gaohong and Gao Xinxing that bet on vehicle-road synergy emphasize the wisdom of the road and focus on the construction of road intelligent infrastructure.
Another category: full-stack enterprises such as Baidu, Mushroom Car Union, Huawei, etc. that adhere to bicycle intelligence + vehicle-road collaboration, emphasizing the attention to the car itself, establishing technical barriers at the three ends of the car road cloud, and laying out the overall operation of smart transportation.
The latter three companies all emphasize the integration of vehicles, roads and clouds, but the technical solutions under the slight distinction also have their own emphases.
Baidu: A fan of the full-stack route. Represented by the smart transportation project of Guangzhou Huangpu District Development Zone, it provides solutions for autonomous driving software research and development- internet of vehicles - intelligent transportation.
Mushroom Car: Late entry, but fast growth. From the strategic level, we seek cooperation with local governments, highly bound to urban public services, and provide a complete set of transportation operation management services through the optimization of operational data algorithms.
Huawei: It's also accelerating. Taking the new smart city construction project in Changsha Wangcheng District as the benchmark, we will build a vehicle-road cloud integration solution from the car end, and provide an ecological layout of "sensor-chip-operating system-algorithm and development application-cloud service".
The war around the city's resources is about to start. Vehicle-road synergy is not only related to industrial reshuffle, but also tests mutual cooperation.
To use an analogy, bicycle intelligence is a soldier in the army, while vehicle-road coordination is more like a commander. According to the explanation of Mo Luyi, executive director of Xiaoma Zhixing, the improvement of the intelligence level of bicycles is to make the soldiers in the army strong enough to deal with various unexpected situations; while the vehicle-road coordination is to coordinate all soldiers together to achieve unified operations.
Looking at the entire combat strategy, the specific can be disassembled into three ends of the road and cloud.
Vehicle end:
Connect cars to the Internet and install smart devices that can receive information provided by the roadside perception system at any time.
End of the road:
The car is equipped with intelligent perception (cameras, millimeter wave radar, lidar, etc.), roadside communication, calculation control facilities and other equipment, and conducts real-time and high-precision monitoring of the surrounding traffic conditions, road obstacles, pedestrian conditions, moving roadblocks, and even road surface leveling, water accumulation and other information.
Cloud:
Provide back-end platforms such as computing and cloud control for automobiles, realize the classification evaluation and execution of road condition information, and deploy the coordination ability between vehicles and roads to the optimal state.
Unlike bicycle intelligence, which relies only on the car end to perceive the outside world, vehicle-road collaboration greatly reduces the information burden that the vehicle itself has to bear.
The combination of smart cars + smart roads + powerful clouds will greatly improve the stability and safety of autonomous driving by allocating complex perception work from the vehicle itself to road systems such as 5G base stations, satellite Internet, sensors and edge computing devices on the road.
At the 2022 China Electric Vehicle 100 Forum, Miao Wei, deputy director of the Economic Committee of the National Committee of the Chinese People's Political Consultative Conference, took the Beijing Yizhuang High-level Autonomous Driving Demonstration Zone as an example and commented that by combining smart cars, smart roads, accurate maps, real-time clouds and reliable networks, it has now become one of the most distinctive pilot demonstration areas in China.
"In the process of realizing the ultimate goal of unmanned driving, there are currently two paths to achieve it, namely the step type represented by Baidu and Waymo and the progressive type represented by Tesla." Miao Wei said that no matter what kind of path, safety is the premise of the development of intelligent networked vehicles, and on this basis, the application of limited scenarios can be moderately accelerated.
03
The lidar that was angrily sprayed by Musk
Can add points to safety
It can be said that the key to determining the outcome of the second half is the intelligent networked car. But now no business dares to say that its car is fully L3. Whether it is Tesla Baidu first, wei Xiaoli, the latecomer, without exception.
Our country divides the automotive driving automation level into 0 to 5 levels, which is similar to the L0-L5 in the CURRENT international SAE classification standard.
L2 is an important dividing line in determining who is responsible for an accident. L2 and below are driver assistance that requires human supervision. The driver remains responsible. At L3 and above, accidents that occur in the state of automatic driving should be determined by the driver or the system development unit.
In this way, on the road to developing higher-level autonomous driving technology, lidar is naturally pushed to the forefront of attacking the level of automatic driving above L3.
Note that the perception device lidar here, as one of the three core technologies of automatic driving, the major genres often win for it.
Musk mocked lidar as a fool's errand, and anyone who used lidar was doomed to failure. Google satirizes Musk's business regardless of whether passengers are dead or alive.
Which is better or worse? In fact, it is not possible to generalize. Learn about the three core technologies of autonomous driving, in addition to the perception just mentioned, there is also planning and control.
Specific to the driving scene, perception is like a person's facial features to feel the surrounding environment, planning is equivalent to using the brain to process road information and make driving decisions, and control can be understood as commanding people to complete the driving.
Cars "don't look good", and there is no way to talk about safe driving. Therefore, perception is often more controversial than manipulating planning. Every business has a different solution, and to sum up, there are two main factions.
Visual Dominance Route:
Represented by Tesla, Moblieye and Baidu Apollo Lite: emphasizing the dominance of cameras, with millimeter-wave radar and advanced computer vision algorithms to complete fully autonomous driving.
Lidar dominant route:
Represented by Google Waymo, Baidu Apollo robotaxi, Wenyuan Zhixing and other enterprises: emphasizing the dominance of lidar, equipped with millimeter-wave radar, ultrasonic sensors and cameras, to achieve long-distance all-round detection.
Visualism is similar to human driving, and the video data obtained by the camera is closer to the real world seen by the human eye, without the need for expensive lidar and high-precision map positioning.
The advantages of ultrasonic radar, millimeter-wave radar and visual algorithms are universal, low hardware cost, but the shortcomings are also obvious, can not solve the long tail effect, the safety factor is not high enough, the camera can not capture the road information at a relatively long distance, nor can it perceive the obstruction and blind spots.
The "radar + vision" laser solution will be much safer. Domestic auto companies such as Xiaopeng Motors, Weilai, and Baidu Apollo have adopted multiple means to integrate to help ADAS capabilities, with multi-sensor comprehensive judgment such as lidar + high-precision map + camera to ensure safety.
Having enough confidence in multi-sensor fusion algorithms may be the reason why the Google faction with weak visual algorithm capabilities still insists on beating Tesla.
The lidar industry has a market of tens of billions, and neither traditional car companies nor technology companies can easily give up. It is foreseeable that when the price of lidar falls, the advantages of all-round detection capabilities and not being susceptible to environmental influences will gradually emerge.
"90% of accidents are caused by human factors, and autonomous driving is to eliminate unsafe driving behavior." In particular, in the future, where driving safety requirements will only become higher and higher, in theory, lidar may become a must-choose choice for self-driving cars.
04
Thinking of giant finance
In the name of "laying eggs along the way"
Do "security intelligence"
The premise of landing and profitability is unmanned and large-scale, and scale requires upfront investment to produce automatic driving.
It's like an endless loop. In the entire field of autonomous driving, Robotaxi's technical difficulty is the highest. Therefore, many companies are exploring the "lay eggs along the way" model on the Road to Robotaxi.
For example, assisted driving with lower steering difficulty provideSAE with L2 and L3 level autonomous driving solutions; for example, it attaches importance to the accumulation of automatic driving data, increases investment in research and development of unmanned trucks, lays out Mini Robobus, and enters the fields of truck transportation, logistics, and personal vehicles.
Compared with the autonomous driving of freight, the commercial landing of driverless taxis is indeed much more difficult. Although the domestic leader Baidu's RoboTaxi began to charge fees, it was also planned to make a profit after 5 years.
From this point of view, laying eggs along the way to climb Mount Everest seems more like a helpless move. If autonomous driving still needs at least a decade of continuous investment from mature technology, commercial maturity to regulatory maturity, then how to successfully survive the pre-dawn darkness has become the most urgent problem.
Chen Qingtai, chairman of the China Electric Vehicle 100 Association, mentioned at the 2022 China Electric Vehicle 100 Forum, "In the revolution of automobile intelligence, the mainland automobile industry has achieved a first-mover effect in changing lanes, but the window of opportunity will not be too long." ”
The value of autonomous driving to mobility is unquestionable, and it is not difficult to foresee that the more complex the technical requirements in the future, the more challenges there will be. But no matter how the industry evolves, cost reduction and efficiency increase are not the ultimate goal. No matter how the company "curves profits", safety is the "first day" of autonomous driving design.
Some references:
Yiou Think Tank, 2021-2022 China Autonomous Driving Industry In-depth Analysis and Outlook Report
Yiou Think Tank, Software-Defined Data Driven, 2021 China Intelligent Driving Core Software Industry Research Report
China Automotive News, "Robotaxi commercialization still has a long way to go, currently facing three major "roadblocks""
Talking about AI, "Pony Zhixing, Wenyuan Zhixing, Momenta's "Post-Bubble Era" Track"
Leifeng Network, "Robotaxi Short Soldier Contact: Baidu to the Left, Didi to the Right"
Finance and Economics, Issue 11, 2021. Into the second half of autonomous driving》
New Intelligent Driving, "Lou Tiancheng Interview: The Value of Autonomous Driving, Only see the tip of the iceberg today"
Xin Zhiyuan, "Tesla can't figure it out, is vehicle-road collaboration the ultimate solution for unmanned driving?" 》
Tiger Sniff APP, "Car Eyes Are Not Good, Let Alone Intelligent Driving"
Gron, "Grasping the Lifeblood of Smart Cars"
Discovery report, "Where has autonomous driving progressed?" 》
"Giant Finance" is produced with heart
Character Observation
Company Observations
Industry Observation
17. Wake up, the metaverse is not a liar
Urban Observations
Digital Observation
Synchronize updates to the following platforms
Today's headlines 丨 Hundred Houses 丨 Snowball 丨 Sina Weibo
Sohu number 丨 penguin number 丨 quick pass number 丨 Zhao Cai number 丨 Fortune number
Gelonghui 丨 Surging News 丨 Ebang Power 丨 Phoenix Network
Chuangyebang 丨 Kaidi Network 丨 NetEase Number 丨36kr
Tiger Securities 丨 Gale 丨 Tiger Sniff 丨 Titanium Media
"Giant Finance" - to create the hottest community matrix in the Internet industry. At present, 21 precision communities have been formed, bringing together more than 8,000 founders, investors and practitioners in the Internet industry, all of which are 100% real-name. Please add the giant financial operation partner @ Zicong WeChat, real-name verification together into the group exchange and learning.
Long press to scan the code to add friends