This novelty has attracted the attention of many people, including automobile media, industry practitioners, college students, geeks and technology bloggers. Several sites on the test ride website are all waiting in line, and the incoming electricity on the first day exceeded 2600. Some people even waited for hours without calling a taxi.
Presumably, everyone wants to know how far Baidu's so-called "autonomous driving" has developed, and want to see for themselves whether the fantasy RoboTaxi has become a reality. However, although Baidu is known as Apollo? Go is fully open, but this "comprehensive" still has many restrictions-
First of all, the operation period is limited, from Monday to Sunday 10:00- 16:00, avoiding the morning and evening rush hours and night driving environment; Secondly, vehicles have preset routes and can only get on and off at the planned 15 stop (similar to car ferry); Thirdly, the speed will be limited below 60km/h; Only 1-2 passengers can ride in the back seat of each car, and the passengers are between 18-60 years old. Most importantly, every car is equipped with a security officer who is ready to take over the vehicle at any time.
So-called super L3 grabs L4 autopilot, but it gives so many restrictions. Apollo of Baidu? How far can you go? How far has Baidu gone in the field of autonomous driving?
To what extent has Baidu's autonomous driving developed?
Baidu's autopilot project started as early as 20 13, and by 20 15, it has invested a total of 20 billion yuan. In 20 17, it broke into the public's sight with a ticket for the Fifth Ring Road in Beijing. Before the road test landed in Beijing, its driverless taxis had been put into trial operation in Changsha and Cangzhou, with a total of more than 654.38+million passengers.
Just one month ago in September 15, a driverless taxi without a security officer carried Li Zhenyu, vice president of Baidu Group and general manager of intelligent driving business group, and CCTV reporters in Beijing Shougang Park.
On the same day, Baidu CEO Li Yanhong predicted the timetable for the commercialization of autonomous driving at Baidu World 2020 Conference. "In five years, self-driving cars will enter the full-scale commercial stage", "With the intelligent transportation system, the traffic efficiency can be improved 15%-30%" and "in five years, China's first tier cities will no longer need to restrict traffic, 10 year".
Behind the ambition, there is indeed enough research and development strength and experience support.
On February 27th this year, in the report card of 20 19 self-driving road test published by California Administration DMV, MPI value (miles? Per？ Among the top five companies involved, three are from China (Baidu, Auto? X, Ma Xiao Zhixing), led by Baidu, surpassed Waymo for the first time in the United States, ranking first.
Besides, in Navigant? In the competitiveness ranking list developed by Research for autonomous driving companies, Baidu also entered the "leader" rating for the first time, at the same level as Waymo and Cruise.
It can be said that Baidu's research and development strength in the field of autonomous driving is not only among the best in China, but also recognized in the global industry.
How to realize Apollo technology?
In terms of hardware, an Apollo test vehicle includes IPC (including special GPU), GPS positioning system, IMU inertial system, CAN bus interface card and large-capacity hard disk (database). In addition to the hardware of these infrastructures, sensors also include top rotating lidar (128 line), forward camera (binocular), side camera (monocular), head/tail lidar (16 line), body left and right lidar (16 line), head/tail millimeter wave radar, and so on.
On the software level, Apollo customized the Linus kernel (4.4.32), but did not list the specific application software and database. As for map positioning, Apollo adopts Novotel's GPS and IMU combined positioning system, which can provide positioning and attitude information with sufficient frequency and accuracy by using Kalman filtering mechanism, which is also one of the best positioning technologies at present.
At the algorithmic level, Apollo joined the DeepDrive Deep Learning Autopilot Industry Alliance and released Apollo? Autopilot data set. Through massive and high-quality real data, the algorithm development and testing of autonomous driving are iteratively updated. Baidu claims that the data volume of ApolloScape is more than 10 times that of similar data sets in China.
The combination of software and hardware constitutes Apollo's autopilot system. Among them, the core of the whole system is IPC, and Neousys? 6 108GC industrial computer supports Nvidia's GTX 1080 graphics card. IPC receives the data sent by various sensors through USB and Ethernet interfaces, and finally drives the vehicle through PCI interface.
Because Baidu is an internet company, it does not have the technology and ability to produce vehicles. Therefore, the carrier of this driver assistance system is Lincoln's MKZ.
Why is this mode? Here's a brief explanation.
First of all, Lincoln MKZ's electrification structure is relatively perfect, with throttle by wire, brake by wire and steering by wire system. Secondly, there is a Dataspeed company, whose main product is ADAS toolkit, which cracks the CAN bus protocol of Lincoln MKZ and packages it into ADAS? Toolkit provided to developers; Furthermore, Ford is willing to provide relevant interfaces for autonomous driving companies, with the aim of expanding travel service providers, which is one of the few choices.
Based on the above factors, it is the most convenient and affordable choice to convert Lincoln MKZ into a self-driving test vehicle. Except Apollo, NVIDIA and Pony? Autopilot companies such as AI also adopted Lincoln MKZ.
Generally speaking, with these technologies, Apollo has been able to achieve L3 level of structured roads, or L3 automatic driving in specific scenes and limited scenes. According to the feedback from the experiencers in Beijing, in most cases, when turning, changing lanes and turning around, Apollo can successfully complete it without the manual operation of the security officer; If the speed of the vehicle ahead is too slow, Apollo will also change lanes to overtake; When pedestrians suddenly appear in the middle of the road, Apollo can take the initiative to slow down and wait for pedestrians to pass before driving.
However, there are also some shortcomings. Because the "habit" of AI driving is not humanized enough, the operation is very blunt when turning and avoiding pedestrians, and it is easy to bump when forcibly turning or braking suddenly. Some passengers said they were slightly carsick.
It can be said that the technical strength of Baidu Apollo is still there, but there is still a lot of room for improvement. So what is the level of car companies in the top position in the field of autonomous driving?
Progress of other competitors
At present, the development enterprises of autonomous driving technology are mainly divided into three categories. The first category is traditional car companies or auto parts manufacturers, including GM, BMW, Audi, China, Bosch and so on. The second category is Internet technology giants, including Google, Baidu, Didi, Ali and so on. Among them, Didi is also a travel platform, which has a dual identity in the development and use of autonomous driving; The third category is entrepreneurial technology companies, including automobiles? X, Ma Xiao knowing and doing, etc.
Here we give two typical examples, one is Tesla, the representative of a car company, and the other is Waymo of Google. And two examples are just the representatives of two technical routes: lidar and visual recognition.
Let's start with Waymo Waymo is a self-driving car program started by Google on June 5438+ 10, 2009, and then it became independent from Google on April 38+06, 2065 and became a subsidiary of Alphabet.
Just earlier this year, Waymo announced that its self-driving road test exceeded 20 million miles; In March, Waymo announced that it had won the first round of external investment of * * * 2.25 billion US dollars, and its pre-investment valuation reached 654.38+005 billion US dollars, far exceeding many car companies such as Volkswagen, Daimler and General Motors.
Compare Baidu's Apollo? Go ahead, RoboTaxi of Waymo was earlier, and it was launched in Phoenix, USA at the end of 20 18. A service, which is also the beginning of Robotaxi's global commercialization attempt. In June, 5438+10, Waymo announced that it would drive in Phoenix in a completely unmanned state (without a security officer) and provide charging service to the public, which marked the first step of Robotaxi's full technical and commercial realization.
Because it doesn't produce cars, Waymo's biggest advantage focuses on algorithms. It is a solution based on lidar, adhering to the business model of "traffic as a service", emphasizing lidar and high-definition maps. Its advantage is that it can quickly build a system prototype, which is not only less dependent on big data, but also easier to detect, extract and divide data, with higher security. But the disadvantages are high cost, weak expansibility and slow commercialization speed.
In contrast, Tesla's Autopilot represents the service mode of "car is product", which gradually transitions from assisted driving to fully automatic driving, with emphasis on camera, visual recognition and light map. Because of the need of visual learning, Tesla relies more on data. To this end, Tesla also has its own "unique cheats"-shadow mode (which claims to have 65.438+000 billion miles of measured data), with lower cost, stronger expansibility and faster commercialization. The disadvantage is that security control is not as good as lidar.
Technically, Tesla is the only car company that independently develops software algorithms and hardware architecture, and its software and hardware technologies have always been in the leading position in the industry. Nikkei BP dismantled model? At 3 o'clock, it is concluded that Tesla is ahead of its competitors in electronic architecture for 6 years; And Thomas, a member of the Volkswagen Board of Directors? Ulbrich also admitted that Tesla was ahead of Volkswagen 10 years in electric vehicles and software development.
The main advantages of Tesla Autopilot are neural network, massive data and control algorithm. Although it is difficult to open the gap with Waymo or Cruise in theory and technology, it has profound experience in the combination of algorithm and vehicle control. It is worth noting that Tesla's strategy of putting autonomous driving into the market is more radical. Just this month, Tesla announced the launch of the fully automatic driving version of FSD? Give a small number of target users a beta version, and realize automatic driving close to L4 level.
From the measured experience, FSD? Beta can do "zero intervention driving" most of the time. It can identify traffic lights and no-go signs at intersections, and automatically select lanes according to intersection markings and navigation. When passing through complex intersections such as roundabout, FSD? Beta can also abide by the rules of intersection concession and avoid pedestrians and non-motor vehicles on the roadside.
At this stage, the narrow road without any marks is a nightmare for all other driver assistance systems, but the camera-led FSD? Beta is still competent, and road boundaries, traffic paths and vehicles parked on both sides can be detected. Even in a parking lot with cars on both sides, FSD? Beta can still identify almost all traffic participants. In the night test, the fire department? Beta's recognition ability is roughly the same as during the day, which is quite accurate.
How far are we from real unmanned driving?
Generally speaking, Baidu Apollo has been in a leading position in China and is in the first echelon in the world. But there is still a long way to go from Waymo, Cruise and Tesla's Autopilot, which is also the goal of Baidu Apollo.
So how far are we from unmanned driving?
It depends on when the difficulties ahead are solved. For example, the technology corner case (corner? Case). The data collected by the vehicle through radar or camera is uploaded to the machine for learning, but in actual driving, it is inevitable that there will be some road conditions beyond the experience of the machine. These are corner cases (such as the Taiwan Province model? 3 example of hitting a truck). According to the research report of the Electric Vehicle 100 Committee, today's driverless technology can handle 90% of normal road conditions, but the remaining 65,438+00% corners have a huge impact and need to be solved within 90% of the time.
For example, the attribution of legal responsibility. The subject of responsibility is a crucial concept in any law. However, autonomous driving technology blurs the division of this concept. The driver is responsible for the unfortunate accident of self-driving vehicles? Is it a technology supplier? Or the brand of the vehicle? These are all unsolved problems (such as the example of Uber causing pedestrian death in the United States).
For example, the right of way and the formulation of road rules. Do unmanned vehicles enjoy the same right of way and accept unified management as manual vehicles? Do they drive in the same lane and apply the same traffic rules?
For example, changes in the nature of products. Automatic driving will greatly improve the utilization rate of vehicles, thus reducing the stock of vehicles on the whole road, because at that time people no longer need the ownership of vehicles, but only the right to use them. Can the public accept this change in nature?
Such as technical ethics. The famous tram problem will reappear. Suppose a self-driving vehicle can avoid the operation of protecting pedestrians when facing the pedestrians who suddenly rush out of the roadside in front, but it will sacrifice the safety of passengers and other vehicles on the road, otherwise it will hurt pedestrians. How should AI judge and choose?
There are still many problems to be faced in the development of autonomous driving, so I won't list them here. Because of this, enterprises committed to autonomous driving technology should not be too hasty, and some unrealized function points will be publicized in advance, which may easily lead to public misunderstanding and even lead to major safety accidents.
China, a 20 13 automobile supplier, conducted a survey on autonomous driving. The results show that 66% of Americans think that "self-driving cars scare me" and 50% think that "this technology cannot run reliably". By 20 18, the data of the two survey results increased to 77%. The reason may be that the traffic accidents of Tesla, Uber and other companies in the autopilot test have affected the public's confidence in autopilot.
The future of Baidu, the future of autonomous driving.
At the Baidu World Congress in 2020, Li Yanhong predicted that after five years, autonomous driving will be fully commercialized, and urban congestion will be greatly alleviated, and there is no need to limit purchases or restrictions. With the gradual popularization of driverless cars, the incidence of traffic accidents will be greatly reduced. Intelligent transportation infrastructure construction based on vehicle-road coordination will improve traffic efficiency 15%-30%, thus contributing 2.4%-4.8% to the absolute growth of GDP.
This business expectation has also been supported by the market, and the business prospect of RoboTaxi has been optimistic by many people and institutions in the industry. McKinsey & Company predicts that China will be the largest autonomous driving market in the world. By 2030, the total sales of self-driving cars will reach 230 billion US dollars, and the travel service orders based on self-driving will reach 260 billion US dollars.
Market research organization iResearch predicts that with the maturity of core technologies such as artificial intelligence, big data, cloud computing and 5G, autonomous driving is welcoming a rapid development stage. By 2022, the global autopilot penetration rate will reach more than 50%, and in 2030, the global autopilot penetration rate will increase to 70%.
No one will doubt the future of autonomous driving. Waymo told all players with a valuation of 1000 billion that autonomous driving is a huge cake. However, the long journey of autonomous driving requires capital, time and technology, and because of the high threshold, it is not like the new forces of making cars, but also a contest between real masters in the industry.
Although there is a long tail effect in technology, the relevant policies and regulations are still unclear, and the year of large-scale commercialization is vague, as the most concerned application scenario of L4 autonomous driving, RoboTaxi has become a popular track.
Baidu, Waymo, Tesla and Cruise, who will dominate this war? ?
This article comes from car home, the author of the car manufacturer, and does not represent car home's position.