Standing in 2020 and looking back at 2019, everyone in the field of autonomous driving can be called a front-line "worker". In order to achieve effective sensing distance of 1,000 meters on the highway, they have gone through countless Modification of the algorithm; they need to stay in the parking lot for several nights in order to successfully park with an accuracy of less than 10cm, or they may try to let go of their hands for the first time so that L4 level autonomous driving can run on the streets... We don’t know Their names, but we remember their teams: Tucson, Momenta, Baidu Apollo, and many, many more.
In 2019, this industry showed its cruel side: investment decreased, accidents continued, and technical routes were denied. But the prospects are also unquestionable. The market size will reach 7.03 billion US dollars in 2021, and China will become the largest driverless market. 2020 is a year of transition, but also a year of precipitation and rebirth.
Article丨AutoR Smart Driving? Nuoyi
The autonomous driving industry in 2019 will have mixed results.
Since 2016, autonomous driving companies have been springing up like mushrooms after a rain. In a red sea, entrepreneurs and large companies are constantly pursuing commercial breakthroughs.
Three years have passed. After financing, integration, and technology accumulation, some of the leading companies have gradually begun to take the technology-first route. In the words of Zhang Dezhao, CEO of Zhidonghe, "Autonomous driving is about to Entering the commercial implementation period, if you don’t immerse yourself in technology research and development, you will easily be eliminated.”
In fact, it is not difficult to see that starting from the second half of 2019, many autonomous driving companies have reduced their external publicity. Large-scale media reports have also turned into small-scale media communication meetings. The good news is that we have seen that many companies in the field of autonomous driving have implemented commercial trials.
In the last two days of December, Baidu Apollo took the lead in obtaining 40 autonomous driving manned test licenses, becoming the first domestic company to launch autonomous driving manned testing in Beijing. Uisee Technology was at the airport Completed the first unmanned freight shipment from the city terminal to the baggage center.
For the first time, Audi China conducted a passenger car platooning L4 autonomous driving and vehicle-road collaboration demonstration in an actual domestic highway scene. The Chinese startup company Horizon provided China's first self-developed L4 autonomous vehicle. level chip - Zhengcheng II, and the autonomous driving computing platform Matrix based on this chip.
Behind these successes is the continuous accumulation of technology and the process of industrial integration. China has become the source of the development of many autonomous driving companies. With the proposal of the "14th Five-Year Plan", the Chinese government will focus on autonomous driving Automobile support is doubled to bridge the technology gap with areas with developed autonomous driving. At the China International Information and Communications Technology Exhibition, the Ministry of Industry and Information Technology, together with China Mobile, China Unicom, China Telecom, and China Tower, jointly announced the official commercial use of 5G, which marks that 5G communication technology is gradually connecting the vehicle end and the road end.
At this year’s World Intelligent Connected Vehicles Conference Summit Forum, Huawei’s rotating chairman Xu Zhijun made it clear that Huawei will use its optoelectronic technology to develop lidar to solve the cost and performance problems faced by lidar.
He said, "Huawei is not a telecommunications company, nor will it become an AI company. Its self-driving chip MDC? 610, which meets the needs of automotive grade, will be released next year. He also bluntly stated Huawei's core manufacturing advantages: Lack of money, simple decision-making.”
The two companies that have to be mentioned in the field of autonomous driving chips are NVIDIA and Intel. As a company that started with graphics cards, NVIDIA has great advantages in AI technology, development platforms and chips. Its technical advantages in other aspects have become a leading company that is difficult to bypass in the field of autonomous driving.
During the GTC China conference that ended last month, NVIDIA not only open sourced NVIDIA DRIVE autonomous vehicle development deep neural networks to the transportation industry, but also released highly advanced neural networks for autonomous driving and robotics. Software-defined platform—NVIDIA? DRIVE? AGX? Orin.
Huang Renxun said: "AI self-driving cars are software-defined cars that must be based on a large number of data sets to drive around the world. We open source our deep neural networks to self-driving car developers and provide them with Provide advanced learning tools that enable them to optimize these networks based on different data sets.
In this way, we are enabling shared learning across enterprises and countries, protecting data ownership and privacy, and ultimately accelerating the rollout of autonomous vehicles worldwide. ”
Intel, as a chip company, has begun to transform into a data company, and in October this year, Intel’s “data-centric” business revenue has exceeded that of “PC-centric” business in the last quarter. "data" business revenue is flat.
In other words, in the past, most of Intel's revenue came from PC-based business, but now at least half of it comes from "data-centric" business. It can be seen that , Intel’s data center business accounts for nearly half of it in a short period of time. If it develops at this rate, its data center business will soon exceed that of PCs.
Intel predicts that the market size faced by Intel in the future will be US$300 billion. With the current revenue progress, Intel still has at least US$230 billion in room for development.
For Intel, there is another big profitable area in Mobileye, a company they acquired. After the just-concluded Mobileye. At the investor summit, Amnon Shashua, president and CEO of Intel subsidiary Mobileye, predicted that Mobileye's revenue will achieve significant and sustained growth in the next ten years. Since 2008, Mobileye will have shipped more than 50 million by the end of 2019. Block’s EyeQ? chip, together with 27 OEM (original equipment manufacturer) partners, provides support for ADAS systems in a total of about 300 models.
He also predicted, “By 2030, ADAS and data will be the key. The potential market size of autonomous taxis is expected to reach US$72.5 billion, and the potential market size of self-driving taxis is expected to reach US$160 billion. ”
The market potential of self-driving taxis is huge.
However, the objective potential market also means that the entry barriers and industry ceilings are extremely high.
p>
Peng Jun, CEO of autonomous driving company Pony.ai, said when his company received China’s highest autonomous driving valuation of US$1.7 billion, “The cake is big enough and cannot be eaten by one company. Therefore, there is no so-called competition in the autonomous driving industry. What is more important is how to work with partners to achieve autonomous driving first. ”
The initial prototype of global Robotaxi driverless taxi travel appeared in December 2018. Waymo launched the self-driving travel service Waymo One in Phoenix, Arizona, USA. In October of the following year, Waymo announced that it would take over the United States RoboTaxi safety officer in Phoenix, Arizona.
After Waymo, Pony.ai took the lead in launching the Robotaxi service PonyPilot in Nansha District, Guangzhou. Baidu operates Robotaxi self-driving fleets in Changsha, Cangzhou and other places. Didi announced that it will A Robotaxi trial operation was launched in Jiading, Shanghai. WeRide started the Robotaxi project in cooperation with a taxi company in Guangzhou in December
Compared with the radical Waymo, at this stage, domestic Robotaxi autonomous vehicles must be operated. There must be a "safety officer" sitting in the main driving position. Tao Ji, Baidu's autonomous driving technical director, once told Zhijiajun, "Baidu Robotaxi safety officers will exist for a long time. ”
Of course, Waymo has an obvious advantage in terms of the number of fleets and data accumulation. In addition, in terms of venture capital financing, Waymo is the most valuable RoboTaxi company in the world.
However, it is foreseeable that , RoboTaxi’s operating projects and fleets with safety officers will develop rapidly in large and medium-sized cities in China by 2020, because this is an effective way to verify autonomous driving algorithms and collect road data.
Standing in 2020, The next ten years of autonomous driving will undoubtedly be a decade of rapid development. During this period, there will be a multi-industry integration effect, and many companies will participate. Of course, many autonomous driving players will withdraw due to capital.
According to data, by 2021, the global driverless car market is expected to reach US$7.03 billion. By 2035, global driverless car sales are expected to reach 21 million units, and China is expected to become the largest driverless market.
Reviewing the Year of Autonomous Driving
You will know who is strong only after trying the autonomous cars running on the road
In the past few years, you have tried smart driving I have ridden in no less than 10 self-driving cars, including car companies, self-driving start-ups and Internet companies. They use different self-driving solutions and have different test ride experiences. Some can shuttle freely in urban areas, while others appear to be unreliable. Too conservative.
The representative companies in China are Baidu and Changan. In May 2018, Zhijiajun took the first test drive of the Xinxiang car with L4 level autonomous driving launched by Baidu and Panda. .
The car all uses the Valet Parking product provided by the Apollo open platform. This product has the characteristics of low cost, wide application, high safety and good interactivity. It also uses 6 cameras and 12 ultrasonic sensors. Radar can realize a complete set of sensor solutions, reducing hardware costs.
For safety reasons, this test ride was arranged in the Internet Industrial Park of Liangjiang New District, which is an office park. The reason is relatively low. During the entire autonomous driving process, the vehicle experienced interference from a large number of pedestrians and vehicles. , speed bumps and downhill sections.
The most impressive thing is that the self-driving car will not drive very fast this time. It will drive automatically at a speed of about 10km/h. When it encounters a pedestrian in front, the vehicle will slow down and stop within a short distance. The vehicle stopped stably at a position about 2 meters away from the pedestrian, and the vehicle continued to drive forward after confirming that there was no one around.
When encountering a speed bump section in the park, the vehicle will automatically slow down and pass. When encountering a downhill section with a fork in the road, the vehicle will choose to stop for a few seconds before continuing to drive after confirming that there is no one or a car. , the vehicle can achieve minimum driving safety while driving.
A year later, the test drive of Baidu's self-driving car was again in Cangzhou. This time, Baidu's self-driving car was already a pre-installed mass-produced model, named Baidu Robotaxi.
The self-driving car is L4 level self-driving and equipped with a Hesai 40-line lidar, 2 four-line lidars, 9 cameras, 9 ultrasonic radars and 2 millimeter wave radars.
Relying on close cooperation with FAW Hongqi, the Robotaxi autonomous driving kit installation plan and the electronic and electrical architecture of the entire vehicle have been redesigned.
The autonomous driving module and the original vehicle structure are integrated to the greatest extent in a pre-installed manner, thereby reducing signal interference and easy loosening caused by modification and disassembly.
At the same time, Hongqi EV has achieved all-round upgrades and optimizations of autonomous driving software and hardware, production line pre-installation and mass production capabilities, in-car human-computer interaction, safety and redundancy guarantees, and cloud fleet management.
In the 3-kilometer road test ride, the overall driving performance of Baidu Robotaxi has been greatly improved compared to before. At the same time, the driving on public roads is also more aggressive and smarter.
For example, when changing lanes to overtake, Baidu Robotaxi will first automatically turn on the turn signal, then perform an acceleration action, and change lanes to overtake after reaching a certain distance. The overall lane changing process is also very smooth, and Unlike novice drivers who are hesitant to change lanes.
When turning left on a green light, Baidu Robotaxi will perform a deceleration action, and then turn left at a certain speed according to the surrounding conditions. In order to verify its stability, Zhijiajun placed the central control armrest at the position of the center control armrest before the vehicle turned. A bottle of mineral water was placed. As a result, the mineral water bottle was very stable and did not fall down when the vehicle was turning.
After experiencing the whole process, you will feel that the whole process is quite smooth and comfortable. The maximum speed can reach 57km/h. On the road, there will be no swaying due to overtaking and lane changes, nor will it be caused by random passing vehicles. The vehicle brakes suddenly, causing discomfort to passengers.
It can be said that the Baidu Robotaxi experience has exceeded my expectations, but for safety reasons, its braking force is too strong and not as gentle as human pedaling.
Compared with Baidu, Changan’s L4 level self-driving car is much more conservative. Changan’s self-driving car is equipped with five 16-line lidars, one 4-line lidar, six cameras, an inertial navigation and High-precision maps.
From the perspective of the sensor configuration scheme, the Changan L4 self-driving car tested this year uses the same sensor configuration scheme as the Changan L4 self-driving car tested in the Changan factory park a year ago. In other words, Changan’s previous choice of L4 level autonomous vehicle configuration plan was correct.
Changan Automobile provided a test drive distance of about 4 kilometers. The road conditions include red street light recognition, obstacle avoidance, avoiding pedestrians, turning left at traffic lights, making a U-turn, following cars, going up and down ramps, merging into traffic, etc., and parking. There are 13 core functions including collision avoidance, bus traffic flow lane changing, and bus traffic flow static obstacle avoidance.
The focus here is on Changan Automobile’s performance in avoiding pedestrians, turning left at traffic lights, and exiting a narrow lane on a ramp and entering a wide lane.
In the initial stage of driving, the vehicle encountered two or three pedestrians on the roadside many times. When encountering pedestrians, the vehicle would perform a deceleration action about 5 meters away from the pedestrians and then avoid them. The avoidance angle would be Depending on the distance occupied by pedestrians on the road, in principle, the vehicle will not deviate greatly from the original lane. Changan has done a good job in this regard, and the deceleration is relatively smooth, and neither pedestrians nor drivers and passengers on the road feel any discomfort.
An unexpected situation occurred during the test ride. When the vehicle turned left and waited for the red light, the vehicle ran through the red light before the red light changed to green after stopping. This was due to the self-driving vehicle discovering that When there is a car approaching quickly from behind, the self-driving car will move forward in order to prevent the car behind it from rear-ending. As for which decision-making ability to choose, a lot of verification is needed.
The situation encountered by Changan Automobile’s new generation L4 self-driving car when exiting the narrow lane of the ramp and entering the wide lane is that the vehicle jitters slightly while driving. This is due to the algorithm that combines GPS and lane lines. There is a 20ms delay on it.
This needs to be improved.
Generally speaking, Changan Automobile’s new generation L4 self-driving car is relatively stable during straight-line acceleration and deceleration. During the lane-changing process, the vehicle’s lane-changing logic is also closer to the habits of human drivers, and There will be no rapid acceleration after the vehicle in front leaves the gap. The ride comfort is good and the steering wheel vibrates slightly. However, the vehicle braking sometimes feels abrupt when encountering traffic lights.
Has autonomous parking become a new trend for enterprises?
In the past year, autonomous parking solutions have been extremely popular in the field of autonomous driving. As a branch of the L4 level autonomous driving solution, autonomous parking solutions On the one hand, the parking solution integrates L4 level autonomous driving technology, and on the other hand, due to the particularity of the site, its commercial implementation becomes a reality.
As a technological innovation product in the era of intelligent driving, Baidu Apollo Valet Parking autonomous parking solution uses Baidu’s unique car, cloud, map and factory integrated solution and the advantages of cloud and high-precision maps to realize intelligent parking. The best cost-effectiveness for end-end transformation, car-end Baidu can realize vehicle mid- and near-environment perception, trajectory planning and vehicle control through car-grade sensors, coupled with Baidu Cloud and Baidu's data accumulation experience and big data analysis capabilities.
Baidu high-precision map has a 100% pass rate in many domestic OEM tests, a relative accuracy of 0.1~?0.2 meters, and a redundancy/missing rate of only 0.01%, thus achieving autonomous parking and cruise accuracy. and high safety. At the same time, based on high-precision maps and visual AI, autonomous parking can ensure 10cm precision positioning and cruising.
Compared with expensive lidar mapping, Momenta uses a vision-based solution to achieve automated mapping. This vision solution can be used universally with automatic parking hardware. During the mapping process, it is extracted through deep learning algorithms Visual semantic features use SLAM technology to automatically generate high-precision maps based on semantics. The entire system can automatically build maps on the cloud and on-board, with an accuracy of 10cm.
New Technology’s autonomous parking solution mainly focuses on the capabilities of autonomous pick-up, real-time parking space search, intelligent parking, and intelligent obstacle avoidance. During the process of autonomous parking, it can identify children and places. Subdivided obstacles such as locks, vehicles, and conical barrels are used to maximize vehicle-side intelligence through centimeter-level positioning. It has the characteristics of high robustness, positioning error distance of less than 5 cm, and positioning angle of less than 1 degree.
Compared with the other three companies, iWalker's autonomous valet parking solution mainly emphasizes the AVP solution based on enhanced visual tags, which combines low-cost car-grade hardware solutions with lightweight field-end modifications.
The enhanced vision tag AVP solution adopted by iMoker is equipped with 4 surround-view cameras, 1 front-view camera, 12 ultrasonic radars and 4 millimeter-wave radars on the car side.
As for sensor hardware, these are sensor combinations that can be mass-produced, and the price can be controlled at the thousand yuan level. Some of the L2-level autonomous vehicles have already implemented some hardware standard configurations (including 10 Domestic models worth RMB 10,000).
According to reports, the AVP solution proposed by Intelligent Traveler is based on enhanced visual tags for global road segment planning, high-precision positioning, local path guidance and special road segment semantic information annotation, through enhanced tags and cloud scheduling. Lightweight site upgrade simulates a complete traffic system suitable for any parking lot.
Bankruptcy, layoffs, the other side of the autonomous driving market
The first autonomous driving company to fall in China was RoadStar (Star Technology), a former star project in the autonomous driving field , RoadSta’s collapse stemmed from an “Announcement of Shenzhen Xingxing Technology Co., Ltd. on Handling Zhou Guang’s Disciplinary Violations.” It only took one month from the time the investors decided to withdraw their capital to the filing of arbitration.
The result of the divestment arbitration filed by Roadstar investors showed that the three founders, Tong Xianqiao, Zhou Guang, and Weifang, were jointly and severally liable for the more than 100 million the company spent.
In other words, the three founders have to repay every investor's money spent by the company. If they cannot cancel the arbitration result through litigation and cannot repay, the three founders may On the list of "distrustful persons".
The latest news shows that the whereabouts of the three founders are also clear: Weijian went to a large company; Tong Xianqiao has no final destination yet, and there has been news about starting a business or working in the past; and Zhou Guang, according to the announcement, raised 50 million USD Shenzhen Yuanrong's official statement is that Zhou Guang is an independent consultant.
Before this, the star self-driving car company Drive.ai, which once shined in Silicon Valley and Ng personally participated in the actual operation, also submitted documents for project liquidation. The submitted documents revealed that it will be shut down at the end of the month and layoffs. Dismissed more than half the staff.
In addition, Velodyne, the originator of lidar, has officially decided to lay off its China office, including its direct sales team and some technical support, and restore its sales model from the direct sales model to the "agency model" that has just entered China. , which means that Velodyne is basically not expanding the Chinese market.
Capital is rational and pursues efficiency. There are countless examples in history that tell us that when major changes such as mergers occur in the market, the industry chaos of burning money and expanding volume will soon stop.
After all, capital has passed the previous era when heroes were judged only by traffic volume. Next, profitability is the fundamental factor that determines financing capabilities.
This article comes from the author of Autohome Chejiahao and does not represent the views and positions of Autohome.