Baidu Driverless Cars Run In Wuzhen, Powered By Four Leading Technologies
摘要： "Baidu aims to achieve Level-4 autonomous driving with its driverless cars. I hopped on one for a short trip and I found the driverless car can already perform actions such as overtaking, wheeling, and avoiding. It seems Baidu can indeed achieve commercialization in three years and mass production in five years!" TMTpost's journalist wrote this report after trying Baidu's driverless car.
Baidu’s driverless cars are finally on the road for the 3rd World Internet Conference.
In fact, Baidu’s driverless cars had showed up in the 2nd World Internet Conference last year. But this time, Baidu’s driverless cars are running on the actual roads of Wuzhen, a town in Zhejiang province. Just before the conference, Baidu’s driverless cars had been tested on Wuzhen’s extended Ziye road. In addition to that, base stations have also been set up for Nokia’s Internet of Vehicles on the extended Ziye road. Xinhua News reported that the Ziye road of Wuzhen will be the first pilot road for autonomous driving.
So far, autonomous driving is categorized into five levels.
Wang Jing, Baidu’s senior vice president and general manager of the autonomous driving business, believes that the current three main tendencies in the automobile industry are electric cars, sharing and smart cars, among which smart cars represent driverless cars that will shape the whole industry.
Last December, Baidu’s driverless cars performed, for the first time, a fully autonomous drive in a mixed traffic situation, driving through the city, the lanes and highways. One year has almost passed and Baidu has turned low-end Chery EQs into driverless cars and has them passed the subject one of the driving license exam.
Wang Jing stated once again on the previous Global Innovation Conference that Baidu is going to achieve minor commercialization in the driverless car sector in three years, and mass production in five years (2021). He added that Baidu’s driverless cars would appear on the 3rd World Internet Conference, which is scheduled from November 15th to 19th. The cars will be running on the actual roads of a city.
Wang Jing believes that the current three main tendencies in the automobile industry are electric cars, sharing and smart cars, among which smart cars represent driverless cars that will shape the whole industry.
Compared with traditional driving, driverless technology has the following four advantages:
Firstly, driverless cars are actually safer. In China, there are more than 500 deaths caused by accidents. With driverless cars, the number will be cut by 100 times. Secondly, driverless cars are more efficient than manned cars. Thirdly, driverless cars follow the traffic rules, and therefore traffic control will be easier to achieve. At last, the cost of commuting will be reduced as the labor cost is relieved.
There are many different technologies needed to make a successful autonomous car, like intelligent interconnect, man-machine interaction, system security. The most difficult technologies of all are environmental perception, behavior prediction, planning control, HD mapping, and precise GPS. These are the most difficult yet fundamental technologies that drive the development of autonomous cars. Today Baidu is going for many of these directions at the same time thanks to our previous accomplishments in AI and deep learning.
Baidu is aiming to be the top one in terms of science and technology research index. And so far we have four top one technologies in our disposal. We have made our own HD map. Last year, we just had a 10cm-level map and now we have improved it to centimeter-level. It’s a consensus in the industry that once you have a 10cm-level HD map, then you have the fundamental tool to make autonomous cars.
What follows is Wang Jing' speech transcript, edited by TMTpost:
I hold high respect to all the entrepreneurs and innovators out there. In big companies like Baidu there are very few opportunities to kickstart a new startup. But I still work really hard to bring about innovation. I have been working at Baidu for six and a half years. I founded Baidu’s cloud computing team and Baidu’s big data business, the very first in China. Besides that I also founded Baidu Research Center, which is the backbone of Baidu’s later AI group. And I also set up Baidu Mobile Cloud Business, Mobile Buinsess, and Baidu’s newest Autonomous Drive Business. So I want to tell you that although there aren’t that many opportunities for me to have my own startups, I can still make some innovations with my power in a company, so as to join the innovation wave.
What I want to share with you today is about smart car and driverless car. Why would Baidu, an Internet company join the fight of the auto industry? If you look at the whole auto industry today, you might realize there are three major tendencies. Even just one tendency can turn the industry upside down, let alone three together. The auto industry is definitely going to experience a bumpy road.
The first tendency is electric cars. The e-car sector has already been brewing for many years. What most car makers do today is making cars powered by electricity. The components in e-cars are like one third of the traditional cars. That said, e-cars are much easier to produce. We don’t need an engine, clutch or driver to make an e-car. So this actually brings many opportunities to Chinese companies.
The second tendency is car sharing. Ride sharing platforms like Uber and Didi are bringing profound changes on the car ownership and use model that will shift the course of the future. It’s the most influential and biggest smart car case ever. When I say smart here I refer to the Internet of Vehicles that many people have heard about today. In the future cars will be a mobile terminal. Cars might become our work place, or even where we live. The use cases will be integrated. In the future you no longer need a smartphone on a car. Still, driverless technology is more influential and disruptive. Many companies around the world are pouring money into this sector for they see the fact that driverless technology can fundamentally shake the core of this industry.
Driverless technology has three advantages. Firstly, driverless cars are actually safer. In China, there are more than 500 deaths caused by accidents. With driverless cars, the number will be cut by 100 times. Humans’ natural reflex is limited. It takes about 0.6 second on average for a person to react and brake. And it also takes 0.6 to 1.2 second for the oil hydraulic system to work and brake. As for driverless cars, they can complete the whole process in 0.1 second in the future. Baidu’s driverless can brake to stop in 0.2 second now and we are working to reduce the number to 0.1 second. Baidu’s driverless cars will rely on electric brake system instead of hydraulic brake system in the future. That said, driverless car can react one second faster than humans, and one second can change many things in an accident.
Secondly, driverless cars are more efficient than manned cars. In big cities like Beijing, Shanghai and Guangzhou, it could take you about two hours to commute if there is a traffic jam. While you are stuck in the line of cars you can only listen to some music. You can’t really work or enjoy your life, definitely a waste of time. Many people might not realize that most of the time, you just park your car somewhere and leave it there for a long time. Statistics on the American market show a private car spends 95% of the time in the garage and only 5% on the road running. And during the 5% of the time, 75% of the time the driver drives alone in the car. Just think about it, 95% of the time a car is parked at a parking spot, taking up space. In a city, 23% of the road space is taken up by parking spots. If we can have 23% of the parking spots for open space like roads or even parks, it would greatly boost the efficiency of a city.
Thirdly, driverless cars follow traffic rules. And they can help out traffic control and boost efficiency since in the future cars and communicate with each other, and even the traffic system.
Fourthly, the cost of commuting will be lowered. According to data from Baidu’s driverless cars, we have estimated that riding a driverless car from Baidu is way much cheaper than hopping on a taxi. To put it in perspective, the monthly cost of commuting with a driverless car in Beijing is only one fourth or fifth compared with commuting with a taxi. In second and third-tier cities, taxi fee is lower but we are still around 70% cheaper. So the cost of commuting will be lowered since you don’t have to pay for man labor. It could drastically reduce the cost.
The endeavors Baidu makes today are preparation for the minor commercialization three years later. We are planning to launch driverless car trial run in some cities. In 2021, we are going to mass produce driverless car. Baidu is the earliest company to announce the mass production of driverless car in 2021. Up till now 18 auto makers and Internet companies in the world have announced to mass produce their driverless car in 2021.
What technologies Baidu has that allow the Internet giant to achieve mass production in five years?
About four years ago, we made the decision to invest in the next-gen AI technology. The concept of AI appeared as early as before the 60s. It was in 1956. However, it had been only more of a science fiction and an unsuccessful technology concept in the past 50 years. It wasn’t until ten years ago, in 2006, the launch of deep learning algorithm was developed that AI technology began to gain some traction. But people at that time didn’t realize the potential of deep learning technology, the fact that it could be used to develop AI technologies.
During late 2012 and early 2013, there were only two companies around the globe that had started to apply deep learning algorithm to their search engine. These two pioneers I am talking about now are Google and Baidu. Google and Baidu are the earliest companies that massively applied deep learning algorithm in the companies’ main business. Baidu was the first company to adopt deep learning algorithm in advertising while Google at that time found it “unsuitable”. And at that time I happened to start directing Baidu’s AI department as well as the advertising department, so I decided to give deep learning algorithm a try. Luckily we had very talented engineers with us, and we were able to be the first one to make it happen.
Baidu started the development of driverless car in 2013 because we saw potential in AI, thinking that it could also bring about changes in this sector.
We can actually say there are two major groups of car makers that adopted AI technology today. First we have traditional auto makers from Germany, America, Japan, and South Korea. But they tend to take ADAS as the entry point. It’s a natural move for them as they need to implement some of the assistive functions in their cars to compete with other brands. That’s why they need to make sure their enhanced cars are still affordable after putting sensors on them.
Internet companies actually have an advantage in this being a new comer, since we don’t have the burden of maintaining mass production of cars. That allows us to cut straight into the level-4 driverless car sector, which is officially driverless. In our opinion, there is a huge gap between level-one and level-4. The factor that sets level-4 from level-3 apart is that autonomous cars are equipped with an expensive device called lidar on the car roof instead of cheap sensors.
A lidar, such as the one Baidu uses, costs around ￥700,000 last year. Today we can buy a 64E lidar at ￥500,000. If you take a look at the two driverless cars outside you would see each one of them is equipped with a 64E lidar that’s priced at ￥500,000 and three 16E lidars priced at ￥100,000. The lidars alone costs ￥800,000, which is too high a cost for most auto makers to have mass production. The data you get from lidars are totally different from those you get from cameras. We can pretty much say that lidar-based autonomous cars are completely different from camera-based autonomous cars at their core.
So today Baidu, Google and Uber are continuing to improve lizards and deep learning technology. What for? We need to get lots of car on the road to gather data, which will be sent to the data center. The amount of data is incredibly large. I am going to tell you a statistic, and you might find it funny. Last year we put out a driverless car on the road for one day. And it took us one week to process and study the data it had gathered to learn driving skills even though there were hundreds of servers in our center. The servers and computers in the data centers are like driving schools. One they learned new stuff they will forward small models to the autonomous cars and enhance its driving ability. This process will continue and make autonomous cars smarter.
Baidu is working on ten technologies relevant to autonomous cars, many of which can actually be the core business of a company. For example, on-board hardware like servers on cars. If Lenovo could bring about good servers that would bring in lots of cash. If you could make a very successful operating system, then you can become the next Microsoft. And there are many different technologies needed here, like intelligent interconnect, man-machine interaction, system security. The most difficult technologies of all are environmental perception, behavior prediction, planning control, HD mapping, and precise GPS. These are the most difficult yet fundamental technologies that drive the development of autonomous cars. Today Baidu is going for many of these directions at the same time thanks to our previous accomplishments in AI and deep learning.
So just how advanced are these technologies that Baidu possesses? If you use this camera to analyze objects, cars, the accuracy is 90.13%, the best in the world. Last year we were able to make it to 89.6%, also the best one in the world. It took us on full year of tears and sweats to make it 0.5% better.
The second best in the world is Google. We are the top two pioneers in this field, there is no doubt of that. But we both believe that cameras are not enough to be the eyes of autonomous cars. Elon Musk, whom I respect a lot is also an active player in this field. But Tesla’s technologies are no match to ours.
When analyzing the pedestrians, the accuracy rate is 95%. As for analyzing the traffic light, the accuracy is 99.9%, which is still not enough. It has to be 100%, and that’s what our engineers have been working on. Baidu is aiming to be the top one in terms of science and technology research index. And so far we have four top one technologies in our disposal. We have made our own HD map. Last year we just had a 10cm-level map and now we have improved it to centimeter-level. It’s a consensus in the industry that once you have a 10cm-level HD map, then you have the fundamental tool to make autonomous cars.
Aside from that, precise GPS is also very important. Today we have GPS that has also reached the centimeter-level, which is a leading technology in China, and the world. Baidu is the only company that has pulled of a large-scale HD map and precise GPS. Other companies are also digging into this field, but they are not making the actual HD maps I am talking about here.
So besides all those stuff I have just mentioned, what other technologies Baidu holds now? I am proud to tell you that Baidu is the earliest company to use GPU to achieve AI. The stock price of the company that had worked with Baidu on this in fact jumped by several times thanks to that. Baidu is the earliest company to use FPGA to develop deep learning technology. This project was highly classified three years ago. We started to dig into this field very early, in secret. And today we are the leading force in the industry. We have the two chips that are most fitted for deep learning technology, GPU and FPGA. Baidu is definitely a leader in these two sectors. Based on that we have also made many innovations on AI algorithms. That led us to driverless driving.
Three years ago, we made a small cluster with 48 servers, with a computing capacity 200% more than the super computer China had back then. This year we are making a cluster for driverless cars, and the final outcome we get actually surpasses Sunway TaihuLight the fastest super computer in the world. Baidu has incredibly hardware computing power.
Baidu’s achievements on AI bring about the accomplishments on driverless cars today. Earlier this year MIT ranked Baidu as the no.2 smartest company in the world. In October, we were ranked as one of the top companies alongside with Google, Facebook, and Microsoft in the AI sector.
In the field of the new generation sensor, Baidu and Ford have jointly invested in the world-leading lidar company Velodyne. That’s the only company that can mass produce 16E lidars so far.
On the second World Internet Conference last year in Wuzhen, Baidu showcased its driverless car, and had drawn chairman Xi’s attention. Chairman Xi actually came to watch the report speech about the autonomous car. In late May this year he came to see Baidu’s autonomous cars too. It helped us gain a lot of publicities. We have also reached strategic cooperation with many companies, auto makers, cities, tourist sights etc., in hopes of getting Chinese driverless cars out in the market.
Last December, Baidu’s driverless car for the first time performed a fully autonomous drive in a mixed traffic situation, driving through the city of Beijing, the lanes and highways, for over ten kilometers. It drove all the way from Baidu’s HQ to the Olympic park. When you were entering this venue you might have noticed that the two cars outside are Chinese brands. Last year it was a BMW, quite high-end. Some asked us if we could make low-end cars go driverless as well, and the answer from us was yes. One year has almost passed and Baidu has turned low-end Chery EQs into driverless cars and has them passed the subject one of the driving license exam. We are hoping to pass another two subjects next year. We want to pass the whole license exam when we finally achieve commercialization.
The previous black car was tested in California. We attainted a drive test license on September 1st so we put it on the road right away to run tests. Trial run in the U.S provides us another set of data that can be compared with the ones we get in China. We need to tell the government how many kilometers the cars have run, and how many times we intervened to convince them it’s safe. It’s very common to do such comparison. So We did ten drive tests in October, and we are satisfied with the result.
And Baidu’s driverless would appear in the 3rd World Internet Conference. This time we are not just putting them out as exhibition, but let them run on actually roads of a city. We will invite passengers to experience our driverless cars.
Driverless cars are actually a common dream for the Chinese auto industry. It’s not something that one company can achieve. It requires the joint efforts of auto makers, the central governments, research institutes, and even banks and insurance companies. Only in this way can we truly make driverless cars available on the open market. We hope that one day we Chinese people can enjoy a better commuting environment while having a lower commuting cost. We look forward a more efficient traffic system. Thank you!
[The article is published and edited with authorization from the author @TMTpost please note source and hyperlink when reproduce.]
Translated by Garrett Lee (Senior Translator at PAGE TO PAGE), working for TMTpost.