Source of this article: Times Finance Author: Xu Dan
Image source: Times Finance
IDG, which has been "raised" by Baidu for 7 years, is about to start earning money to support its own family.
On December 21, according to Tencent News, the employee assets of Baidu Intelligent Driving Business Group (IDG) have been transferred from Baidu Group's relevant main companies to several Baidu wholly-owned subsidiaries this year. After IDG was fully transferred to the new entity in the first quarter of this year, some commercial departments needed to develop and complete annual revenue targets.
Behind Baidu's business adjustment is also the beginning of the commercialization of the entire autonomous driving industry. On November 25, Beijing officially opened the first pilot project for the commercialization of autonomous driving services in China, and Baidu and Xiaoma Zhixing became the first batch of enterprises to be allowed to carry out commercial pilot services.
On the afternoon of November 25, a resident of Yizhuang, Beijing, used Baidu Apollo's self-driving travel service platform "Radish Run" to complete the first payment, and the passenger took a 2.1-kilometer ride and spent 1.06 yuan.
Turning the time back to 3 years ago, 2018 is still a prosperous and ethereal year for the concept of automatic driving, the industry leader Waymo valuation is as high as 100 billion US dollars, and the news of "L4 mass production landing" has also risen and fallen in China, but so far automatic driving has not really landed, and Uber has also made the first case of automatic driving death.
The trough may be followed by the industry's second upward climb. However, for the industry with high technical content of autonomous driving and related to traffic safety, this period of climbing will also be very long. Baidu IDG's dream of making money is still difficult to achieve in the short term.
"There are already a lot of people who use it as a commuting tool"
Yizhuang is located in Daxing District, Beijing, is the only area in Beijing that enjoys the dual preferential policies of national economic and technological development zones and national high-tech industrial parks. It is located in the automobile research and development center of BAIC BJEV, Mercedes-Benz, Xiaomi and other companies, and has the advantage of a natural industrial cluster. In addition, Yizhuang has regular roads and relatively little traffic, making it an ideal test site for autonomous driving.
Yizhuang autonomous driving test area Image source: Times Finance
On the afternoon of December 1, Times Finance came to the automatic driving experience section of Yizhuang Development Zone.
To play self-driving cars, you need to download two company-specific Apps - Radish Run and PonyPilot+, there is no function of getting on and off at any location, passengers need to find the nearest automatic driving site to get on the car, but the two companies have a wide distribution of sites, about 200 Xiaoma Zhixing sites, Baidu is more than 400, can cover the core area of Yizhuang.
The rest of the interface is not much different from the general taxi software, less than a minute after the taxi, the system shows that the order has been received, 10 kilometers away, the normal price is more than 50 yuan, the initial commercialization platform discount is larger, in fact, only need about 2 yuan.
After boarding the car, the screen on the back of the front seat clearly shows the travel route, simulated road conditions and real-time speed, and there is a more complex screen next to the safety officer to show the surrounding road conditions and the driving status of the autonomous driving.
In the process of driving, automatic driving can already be skilled in basic operations such as intersection turns, lane changes, and traffic lights, and in general, there is no obvious difference between manned driving, but automatic driving will be more cautious and more compliant with traffic rules.
The seatback screen shows that the maximum speed of autonomous driving is 60 kilometers per hour, but it is significantly slower than other vehicles on the road. On flat roads with few vehicles and pedestrians, automatic driving can reach the maximum speed, and once it encounters overtaking, congested road conditions, pedestrians crossing the road, etc., automatic driving will brake and slow down. If you take autonomous driving during peak periods with complex road conditions, there will be sharp braking from time to time, because automatic driving needs to avoid vehicles and pedestrians around that do not obey traffic rules.
At present, all autonomous vehicles are equipped with safety officers, and during the experience of Times Finance, the safety officers did not take over the steering wheel the whole time, but they would put their palms up on their legs and be ready to take over at any time.
A safety officer told Times Finance that his current takeover frequency is about once a day, most of which occur when encountering vehicles and pedestrians who do not follow the rules. Although the frequency of takeovers is low, safety officers will concentrate on paying attention to the road conditions on both sides during the driving process.
Times Finance has learned that a prerequisite for becoming a safety officer is more than 10 years of driving experience, but this is not a position for ordinary drivers. Compared with ordinary drivers, safety officers are more like self-driving testers. The above-mentioned safety officer said that he was originally responsible for technology-related work, and the work of the safety officer was not only to ensure the safety of automatic driving, but also to train the automatic driving system based on driving data.
At present, there are nearly 100 autonomous vehicles running in Yizhuang, and this "black technology" has been integrated into the lives of many Yizhuang people. Safety officers say it's been rare to meet people who are riding in self-driving vehicles for the first time recently, "Autonomous driving is now cheap, and many people have chosen to use it as a commuting tool for commuting to and from work." ”
Baidu told Times Finance that since its commercialization in November, the overall satisfaction rate of users riding autonomous driving has been 4.9 points, and the praise rate is greater than 94%.
Investors: The 2C business of autonomous driving is still "self-deprecating"
Taking the Gartner curve as a benchmark, autonomous driving is currently in the stage of slowly climbing upward from the trough period.
AI image recognition, edge computing, AI chips and other autonomous driving related technologies are gradually maturing, lidar is also entering the eve of mass production, the maturity of technology and policy support so that autonomous driving manufacturers began to consider the issue of commercialization.
Xiaoma Zhixing told Times Finance: "Obtaining the commercialization pilot license means that we can open the Robotaxi commercial closed loop for the first time, and the passenger car business of Xiaoma Zhixing has officially entered the commercial model verification stage from the technical normalization test." ”
However, from the current stage of the market situation, automatic driving is still in the stage of burning money, the input cost is extremely high, and the industry has not yet made a profitable case.
The fastest self-driving companies are in the trucking industry. One of the head companies Tucson has been listed in the future, with full-year revenue of 2021 between $5 million and $7 million, but the capital expenditure reached $14-18 million.
Compared with unmanned trucks, taxis such as the Robotaxi are more expensive. Startups mostly raise more than $1 billion, and Robin Li revealed in an interview with Bloomberg that Baidu spent a total of $20 billion on autonomous driving in 2020.
But after taking the first step towards commercialization, companies have already made profit plans. Wang Yunpeng, general manager of Baidu's autonomous driving technology department, publicly calculated an economic account with the media: if Robotaxi costs 500,000 yuan, has 5 years of reliable operation time, the monthly cost is about 8,000 yuan, which is the same as the current online car, according to the average monthly revenue of 20,000 yuan, 2 and a half years can balance the cost of vehicle hardware, and the 5-year operating cycle can make a profit of 700,000 yuan.
Due to the reduction in hardware costs such as lidar, Baidu's latest fifth-generation Robotaxi cost 480,000 yuan, Wang Yunpeng told the media, to the seventh or even eighth generation, the cost can be equal or even lower than private cars. In addition, Baidu's vehicles in Shougang's test operation have removed the safety officer, and the seventh-generation car no longer retains the driving wheel, and next year the official will allow manned tests without steering wheels.
Regarding the pricing of autonomous driving services, Baidu told Times Finance that after the subsidy ends, it will adopt a more market-oriented charging standard, which will refer to the level of online ride-hailing.
According to Robin Li, about 2025, the overall cost of Baidu Robotaxi will be lower than the cost of online ride-hailing, and the scale will expand rapidly to achieve profitability.
Baidu self-driving car driving on the road Image source: Times Finance
But not everyone is so optimistic. Yang Ge, founding partner of Star Han Capital, told Times Finance that from the perspective of industry logic and technical development, he does not think that Robotaxi's business model will land in the short term.
All technology-based industries promoted by B2B cannot develop To C in the short term, and To C is actually just everyone's self-congratulation. ”
"The upstream revenue of the autonomous driving industry comes from the midstream, the revenue of the midstream comes from the downstream, and the downstream revenue comes from the surrounding industries, and the conditions for the outbreak at the C-end have not yet been born." That is to say, in recent years, the development of autonomous driving has been driven by the mutual demand between upstream, middle and downstream industries, rather than the needs of the public. ”
Large-scale landing? The technology is not ready
As a technology related to traffic safety, the "popularization of automatic driving" does not only refer to the maturity of technology, but also contains the comprehensive role of economic, legal, moral and other factors.
"The popularity of autonomous driving is a very difficult thing, and its difficulty can even be compared to cancer treatment." Yang Ge told Times Finance, "It is not difficult to overcome the technology, but difficult in robustness, the ideal state of automatic driving is that no matter how complicated the road conditions are, how to drive will not be wrong." ”
From a technical point of view, improving the stability of the system requires continuous testing, and the test is divided into road test and simulation test. Road testing is the test system based on actual road driving data, which is the most effective but also the most demanding test. According to the RAND think tank, a true autonomous driving system needs to go through at least 11 billion miles (about 17-18 billion kilometers) of road validation to meet the conditions for mass production applications.
This figure means that even if 100 self-driving vehicles are tested 24 hours a day, 24 hours a day, the time it takes to accumulate data is in centennials.
Therefore, most autonomous driving companies will supplement the road test with simulation tests. Simulation testing is the construction of a virtual traffic environment, through the simulation of road conditions training system. But simulation is always a virtual environment, and it can't simulate everything in reality.
"In the social environment, there are various situations such as touching porcelain, plugging, and grabbing roads in the middle, which cannot be done in the simulated environment." Yang Ge said.
Some autonomous driving entrepreneurs told Times Finance that most autonomous driving manufacturers are currently unable to accurately repeat the problems in the real road test scenario in the simulation system. That is, if the automatic driving has an accident during the driving, the accident scene cannot be played back in the simulation system to ensure that the improved system does not make the same mistake.
From the perspective of regulations, at present, the state and society do not have clear responsibility determination standards for autonomous driving accidents, "If automatic driving causes casualties, how to judge responsibility?" Technicians, autonomous driving manufacturers, car owners? Who is responsible? It is likely that the entire project was shut down after an accident. ”
The mutual struggle between technical issues and legal and ethical issues makes the robustness of autonomous driving very difficult. "Society is a complex and unpredictable chaotic synthesis environment, which is completely different from the computing environment. The robustness of autonomous driving must reach the level that social safety and legal ethics can solve, not within 5 years. Yang Ge said.
In addition, each self-driving car needs to independently judge the surrounding information system, and when there are many self-driving cars on the road, there will be interference between different systems.
"This is the same as the single pendulum, double pendulum, and multiple pendulum problems in physics. If there are more than 50 cars gathered at an intersection, each at a different speed, and each car is making a judgment, the chaotic system collapses. You can't use computational subjectivity to complete the interpretation of the chaotic world, in simple terms, as long as there are more than 20 self-driving cars at the intersection, the computation may immediately get stuck. Yang Ge said.
If you want to solve the problem of robustness and chaos from a technical point of view, the only feasible path is "vehicle-road collaboration", which is also the mainstream solution of domestic automatic driving.
Vehicle-to-road collaboration needs to link people, vehicles, roads, and clouds together, transmit real-time data to the cloud, and feedback information to each traffic participant through computational analysis.
However, the biggest problem with vehicle-road collaboration is that the project is too large, it is no longer a single point problem of the popularization of automatic driving technology, but the cross-border integration of various industries such as automobile companies, transportation companies, and communications industries, and large financial support. Its requirements for the productivity of social organizations are very large, and in the next few years, the realization of vehicle-road coordination will be more difficult under the condition that the global economy is still uncertain.
"With normal international relations, social stability, normal capital flows, stable annual growth of the economy, and strong support from the state, I think it will take 10 years to have a city-level demonstration point." Yang Ge said, "It is more difficult to transform old cities, basically only in areas like Yizhuang, because the original buildings are few and the density of residents is low." ”
Yang Ge said that he currently has no intention of investing in autonomous driving projects, "I will look at this industry, and I will not reject any future productive industry, but at this stage, automatic driving is not make sense." ”