Car stuff (public number: chedongxi)
Author | Haohan
Edited by | Xiaohan
This time, Baidu Apollo overtook Waymo.
Today, the Beijing Intelligent Connected Vehicle Policy Pilot Zone officially released the "Unmanned Manned Test Permit" and issued the first batch of domestic autonomous driving unmanned demonstration application notices to Baidu.
This also means that under the permission of the policy, Baidu Radish Run can carry out unmanned manned demonstration applications within the scope specified by Beijing Yizhuang.
It is understood that Baidu Apollo invested 10 unmanned vehicles in the first batch, and the follow-up plan is to add another 30 unmanned vehicles. Today, Cheshi was also invited to experience the Robotaxi of Baidu Apollo's main driving, and the vehicle driving process was very smooth and arrived at the destination safely.
But don't underestimate the change in the position of the safety officer, which is not only the first batch of unmanned manned test vehicles in China to officially start running, but also a milestone in the development of China's automatic driving industry to fully unmanned.
▲ One shot to the end to experience Baidu Apollo main driver unmanned Robotaxi (clip)
Just last month, Baidu Apollo's rival, Waymo, across the ocean, also obtained a commercial license for self-driving taxis from the California Public Utilities Commission (CPUC) and allowed the use of self-driving cars to provide passenger services when equipped with safety officers.
This also means that with the opening of the unmanned autonomous driving travel service by Baidu Apollo, it has already equaled Waymo in the second half of the commercial landing.
But if you think about it carefully, compared with the road conditions in California and other regions of the United States, the domestic road conditions are obviously more complicated, there are more types of traffic participants, and to achieve the same unmanned autonomous driving travel services, Baidu Apollo will face more problems and higher gold content.
Therefore, for Baidu Apollo and even the entire Chinese autonomous driving industry, it is not only a tie, but also a counter-overtaking.
First, China's road conditions are more difficult Baidu is the leader of the anti-superiority
For a long time, Alphabet (Google's parent company) subsidiaries Waymo and Baidu Apollo can be regarded as the two major players in the global autonomous driving industry, and their technical routes, operating fleets, commercial landings and other every move are the focus of the entire autonomous driving industry.
If the development of the automatic driving industry is divided into the first and second halves, then the first half is a technical competition, and the companies are basically "building cars behind closed doors", and the second half has become a time for each business to land and prepare for "homework".
Looking at the commercial landing process of Waymo and Baidu Apollo, in 2018, Waymo launched an autonomous driving manned service, and in 2019, Baidu Apollo's unmanned taxi announced that it would start operations in Changsha, Hunan Province; in October 2021, Waymo obtained a california driverless taxi charging license, and Baidu Apollo also obtained a domestic charging pilot license at the end of last year.
Overall, Waymo benefits from relatively simple road conditions and more open policies, which are slightly faster than Baidu Apollo, but it can also be seen from the interval time change that Baidu Apollo is accelerating.
With Baidu Apollo obtaining the unmanned autonomous driving manned license this time, it can be said that it has equalized Waymo from the commercial landing level.
▲Baidu Apollo opened operations in Changsha
If you look at the complexity of the actual testing and operation of the two companies, compared with Waymo's testing of road conditions in the United States, Baidu Apollo's road conditions in china are obviously more complex.
Specifically, Waymo is mainly engaged in technology research and development and testing in the United States. Its urban core area is very small, the residential and living areas are relatively scattered and sparse, the connections between the main areas are mostly trunk highways and branch roads with relatively low traffic flow, the overall road scene is more standardized, and the traffic participants are relatively small.
China's urban core areas are larger and denser, and living areas, commercial areas, and residential areas are mixed, and congestion is also the norm during peak travel. In addition to motor vehicles, traffic participants also include a large number of pedestrians and non-motorized vehicles, especially various express delivery car takeaway vehicles, which will bring relatively large traffic risks.
▲Pass through traffic lights without traffic lights
Xu Baoqiang, director of Baidu's automotive robot department, told Che Dong that according to Baidu Apollo's actual test data in China and the United States, there is a significant gap in the number of traffic obstacles in the test area of the two places, and the test city is often the most representative area of the country, so it is estimated that the participants and complexity of China's roads are 15 times that of the United States.
It can be seen that although Baidu Apollo has already caught up with Waymo in business, Baidu Apollo faces a more difficult scene and a higher technical content, which to a certain extent surpasses Waymo, the big brother of the industry.
▲The vehicle makes a right turn
The problem here is that at the beginning of April this year, Waymo also opened the Robotaxi service without safety officers in the car to its San Francisco employees, and Baidu Apollo had a safety officer in the car, so is it still behind Waymo?
The answer is no. Baidu Apollo actually let all unmanned vehicles start running earlier.
In October last year, Che dong also experienced Baidu Apollo's fully unmanned Robotaxi in Beijing's Shougang Park. During the ride, the vehicle autonomously completed the U-turn, lane change, unprotected left turn and other operations, the overall driving is stable, and when encountering pedestrians in front of it, illegal parking vehicles, the vehicle can quickly identify, by honking the horn, braking and other actions to complete the autonomous avoidance.
According to Xu Baoqiang, Baidu Apollo can already technically achieve the ability of all people, and it can also run on the road after the policy is released.
▲Vehicles change lanes
So, after Baidu Apollo equalized Waymo, who took the first step in the first step of all-unmanned commercial operation, and now it seems that Baidu Apollo's chances of winning may be greater.
Second, thousands of people research and development for 9 years to overcome a number of technical problems
Since Baidu Apollo's Robotaxi is obviously more complicated than Waymo, how did the Baidu Apollo team get it done?
At this stage, autonomous driving technology can be simply divided into three parts, namely perception, decision planning, and control.
With the rapid iteration of on-board perception hardware, perception algorithms, and the continuous maturity of wire control technology. Therefore, relatively speaking, in the complex and changeable urban road conditions, decision-making and planning have become the most difficult and tangled part of the relative.
▲Baidu Apollo fifth-generation autonomous driving kit
Specifically, perception is the collection of road data through various sensors, and then through the perception algorithm to let the vehicle know what is around it and how far away it is. The control is the control mechanism to control the vehicle forward and backward, turn left and turn right.
Therefore, between perception and control, it is necessary to process the perception information and make driving decisions, and then pass it to the actuator to complete the operation accordingly.
For example, human drivers often hesitate to give way or rush when driving around the island, turn left unprotected, and meet the car, and the same is true for self-driving cars, which can be regarded as the most difficult part of decision planning.
▲Roundabout game puzzle
In this regard, Xu Baoqiang told Che Dong that taking the roundabout scene as an example, more refined environmental information will be constructed for the scene first, such as more accurate obstacle orientation, lamp language, etc., to provide richer input for the downstream.
At the level of decision planning, Baidu Apollo will first abstract the policy expectations from the traffic regulations to ensure the compliance of driving behavior; at the same time, it will use the learning method to determine the super-decision of various interactive obstacles in a data-driven way, and then ensure driving safety through a set of safety frameworks through theoretical calculations; for potential dangerous scenarios, it will further improve safety through defensive driving strategies.
▲Play with the opposite vehicle at the intersection
In this way, the vehicle driving behavior is expected to be controllable and safe and compliant.
At the same time, the vehicle-to-road coordination system also further strengthens the vehicle's over-visual distance perception ability, providing support for solving problems such as ghost probes and occlusion.
▲5G cloud generation cockpit
It is reported that the actual road test mileage of Baidu Apollo has exceeded 27 million kilometers, and after the massive road test data is backhauled, one is to establish a scene set through classification and labeling to help continuously iterate the algorithm; the other is to form a baidu Apllo model training closed loop, with a large amount of data to support typical learning tasks such as perception and prediction.
Then, the reason why Baidu Apollo can solve more problems depends on the thousands of R & D teams and technical reserves behind it.
Baidu has been laying out autonomous driving since 2013, and launched the world's first autonomous driving open platform Apollo in 2017. By the end of March 2022, Baidu's autonomous driving China patent applications had accumulated more than 3,700 publications, ranking first in China.
▲Avoid large vehicles facing each other
All in all, Baidu's unmanned test is by no means as simple as showing muscle. It is precisely because of the systematic technical strength behind it that it has the confidence to open unmanned autonomous driving travel services under the open road and continuously expand the scale of commercial testing.
Third, L4 mass production is not a problem has opened up the upstream and downstream industrial chain
In addition to solving the common long-tail problem in mass production landing, Baidu Apollo also has more advantages than Waymo in terms of vehicle front-loading mass production and landing scale.
From the perspective of passengers, the density of the fleet is not high, there is no way to achieve truly convenient travel; from the perspective of service providers, the size of the fleet is not large, one is that there is no way to accumulate more data to improve the automatic driving system; the other is that it is impossible to achieve large-scale commercialization.
With most autonomous driving companies already entering mass production, if the scale of the fleet is expanded through aftermarket modifications, it is obvious that it cannot keep up with the rhythm of the times.
▲ Baidu Apollo fourth-generation unmanned vehicle
Therefore, the ability to mass-produce unmanned vehicles in front of it has become a major focus of the layout of autonomous driving companies at this stage.
At present, waymo's Robotaxi operated in the United States is basically modified from the Chrysler Grand Jaylon and Jaguar I-PACE addition kits, and there is no real front-loading mass production unmanned vehicle. The unmanned vehicle project it cooperated with Geely is still in the concept stage, and there is still a certain time before mass production.
In contrast, Baidu Apollo reached an in-depth cooperation with FAW as early as 2018, and the two sides jointly designed the baidu Apollo fourth-generation autonomous driving kit with the Hongqi EV model, and produced by FAW's passenger production line car.
In October 2021, Baidu Apollo released the fifth generation of unmanned vehicles, including three models, namely Apollo Moon Polar Fox Edition, Weima Edition, and Ean Edition.
▲From left to right, they are Apollo Moon Aehan version, Polar Fox version, and Weima version
Among them, the Polar Fox edition is jointly created by Baidu Apollo and BEIQI Extreme Fox, and is also Apollo's second front-loading mass-produced unmanned vehicle.
The car is customized based on the BAIC Jihu Alpha T pure electric model, and the automatic driving sensor is configured with 13 cameras, 5 millimeter wave radar, 1 main lidar, and 1 redundant lidar, and the computing power of the computing platform reaches 800TOPS.
In terms of cost, the cost of the Apollo Moon is only 480,000 yuan, which is one-third of the average cost of the industry's L4 level autonomous driving models.
If calculated according to the 5-year operating time, the average monthly cost of Apollo Moon is 8,000 yuan, which is lower than the cost of manual driving in first-tier cities, and the cost of chauffeur-drive is basically the same in second-tier cities.
▲ Apollo Moon Polar Fox Edition
This also allows unmanned vehicles to enter the operating cost range of online ride-hailing, providing a low-cost advantage for its commercial landing.
Of course, in addition to mass production, Baidu Apollo also actively cooperates closely with upstream and downstream enterprises in the industrial chain.
As the most important hardware in the L4-level automatic driving perception system at this stage - lidar, Baidu Apollo has reached a cooperation with hesai technology, a domestic head lidar company, and the two sides have jointly developed lidar.
▲Baidu Apollo's fifth-generation unmanned vehicle custom lidar
In general, in the second half of the entire autonomous driving, Baidu Apollo not only has the technical advantage of continuous iteration, but also has the guarantee of large-scale landing of the industrial chain.
The car thing was learned from Xu Baoqiang, director of Baidu's automotive robot department, that Baidu Apollo's next-generation unmanned vehicle is also on the road, which is worth looking forward to.
Fourth, the total mileage of more than 27 million kilometers covering the super first-tier large cities
In terms of landing, Baidu Apollo and Waymo have won and lost each other.
At present, baidu Apollo fleet size has reached the level of 500 vehicles, while Waymo fleet size is about 800 vehicles. Although Waymo is relatively ahead in this round, by 2023, Baidu Apollo plans to have 3,000 L4 self-driving unmanned vehicles.
Although Baidu is temporarily lagging behind in the size of the team, it is significantly ahead in the number of landing cities.
Today, Waymo mainly operates in Los Angeles, San Francisco, New York and other cities, while Baidu Apollo has fully opened test ride services in Beijing, Changsha, Cangzhou, Guangzhou, Shanghai, Chongqing, Shenzhen, Yangquan and Wuzhen, and has begun to commercialize paid operation services in Beijing, Chongqing and Yangquan.
More cities mean richer scenarios, which allows Baidu Apollo to collect more Corner Cases, which is more advantageous in algorithm iteration.
At the same time, the algorithm developed based on more urban driving data is also stronger in generalization capabilities - that is, after Baidu Apollo's Robotaxi algorithm is stable, its unmanned vehicles can be put into use in new cities without the need for adaptive training (or only a small amount of training).
Algorithms based on a small amount of urban data are difficult to achieve similar effects.
In the future, Radish Run plans to expand its business to 65 cities by 2025 and 100 cities by 2030. In terms of test mileage, Baidu Apollo currently has a total actual road test mileage of more than 27 million kilometers.
For enterprises, getting enough test licenses is the key to the rapid increase in test mileage. Up to now, Baidu has obtained 593 autonomous driving test licenses in China, including 398 manned test licenses.
In summary, whether from the technology/talent reserve or from the landing operation, whether from the policy guidance or the public awareness, the entire China's autonomous driving industry has been at the forefront of the world.
At the same time, with the mainland's dominant 5G cloud driving, vehicle-road collaboration technology system and perfect regulatory system as the support, coupled with the active layout of large Internet companies such as Baidu in automatic driving technology, China has not only gained a first-mover advantage second only to the United States in the autonomous driving track, but also become the only country in the world that can keep pace with the United States.
This time, with Baidu Apollo opening up unmanned autonomous driving travel services, it has surpassed Waymo, the leader in the US autonomous driving industry, and even began to take a step ahead in the field of track segmentation and evaluation standards.