Valeo China CTO? Gu Jianmin
Thank you for your invitation. I am very glad to have this opportunity to share it with all the leaders, experts and colleagues present here. This topic is also very big, "the road from ADAS to autonomous driving." Personally, active security is an extension of passive security intelligence. If we go further, ADAS is what we generally call a driver assistance system, which is an intelligent extension of active safety.
Is autopilot an intelligent extension of ADAS? In a sense, yes, but autonomous driving is not just a technical problem. This morning, two spokesmen said that it also involved the landing of scenes and business models. In addition, it also includes regulations, infrastructure and insurance, which are very related to autonomous driving, so today we are discussing not only technical issues.
Because I speak on behalf of Valeo, I believe many people here know Valeo better. Valeo is an integrated supplier of auto parts, headquartered in Paris, France, and we rank in the top ten in the world. Valeo also has many layouts in China, with 35 factories and 12 R&D centers, as well as a factory and an R&D center in Nanjing.
From the product line, it can be said that if you drive a car, there must be Valeo products or parts in this car. We have four business divisions, and the main product of one business division is the autopilot driving assistance we are going to talk about today. In the product division, there is a sensing system, that is, sensors, laser radar, advanced human-computer interaction of artificial intelligence, and car networking, which is provided to everyone to help them build a car that meets their travel needs. This is a brief introduction of our company in autonomous driving.
If we watch it again, what is our topic today? From ADAS to autonomous driving, so I put this page here, this page PPT, in fact, I used it last year, and I haven't changed almost a word today, because this view has not changed.
In the first sentence, how to do autonomous driving, how to help the commercialization of autonomous driving, what is the first? What is the best way to enter a market? Starting from a young age is to start with simple and low-cost autonomous driving technology. I'm talking about technology here, starting with simple low-cost technology.
What's next? The purpose is to attract enough users who are willing to pay, because we all know what autonomous driving is. Demonstration test, no problem, everyone will welcome it, but you can't land in business yet. What are the basic conditions for commercial landing? Someone needs to pay, there will be no pie in the sky, and someone has to pay, either you or our car factory.
How to do it specifically? Below I list several scenes or commercial landing methods, starting with automatic parking or parking service, because as we all know, parking is slow and the scene is relatively controllable, in a semi-closed parking lot or garage. And start with low-speed automatic driving. The speed listed here is 40 kilometers per hour. In fact, this speed is already very high. Generally speaking, there may be more than 40 kilometers of vehicles on the expressway. What to do first at low speed? It can challenge the perception system and decision-making system with less pressure.
This should start with simple technology.
What else is there? Starting from a specific scene and a specific purpose, there are many scenes of automatic driving, and it is meaningless to drive automatically without escaping from the scene. For an extreme example, if you are in a 300-meter-diameter proving ground, there are no vehicles or obstacles in it, let alone L4 and L5. But in another scene, L3 can't even do it in a very congested situation.
The key is to remove the safety driver. Many of our demonstration and test vehicles today must have safety officers on the road to drive automatically, which is also limited by our current laws and regulations.
But think about it, if there is a security officer, we usually talk about L4 vehicles, or on the basis of L3. If we can't break through this aspect, our technology is still on the technical level of L3, essentially.
Of course, another point discussed today, in fact, the real autopilot should not be entangled in L2, L3 or L4. What I see today is to look at the scene, how to break through commercialization and find a commercialization model. That's all that matters.
Finally, the demand for delivery may be more practical than transporting passengers. Of course, from the perspective of safety, goods may not receive as much attention as guests, which is not the only reason. If you look at the past few months, especially when the epidemic broke out seriously, what will we see in Wuhan and Beijing? Some unmanned logistics vehicles transport medical equipment and medical supplies, which can avoid people-to-people contact, especially in some places with serious epidemics. This is also the demand for scenes where we see that unmanned logistics vehicles may have more demand than giving people away.
This is one reason.
I am here to throw a brick to attract jade and throw out these points.
Next, please allow me to spend some time with Valeo's products to tell you in detail how we found the scene and the ultimate goal of commercialization.
As I said just now, automatic parking is a relatively easy scene to realize. Usually speaking of automatic parking, what is parking assistance? The driver needs to complete automatic parking or parking assistance according to the prompt of the in-vehicle system. However, once the driver enters the car, our customers can choose to park inside or outside the car, which is remote parking.
Valeo launched the remote parking function on 20 16, and it has also been mass-produced. You can have a look. Using the remote control key, you can stop with one button when you need to stop in some emergency situations.
Next, we can go further. We can imagine that if we are in an underground garage, we can use remote parking to make the vehicle stop automatically, which is similar to the remote parking technology just now, except that the vehicle may need to travel a greater distance or find a parking space; The second difference is that what we are talking about here is parking service, which needs the support of the factory. In terms of industry, there are two trends or two methods. One is that the parking service is completely completed by the sensor at the vehicle end, and the other is that the factory end and the vehicle end need to cooperate to complete the parking service.
If you rely on the sensor at the rear of the car, it may take a long time to find a parking space in a very crowded underground garage, and it may also cause parking congestion. Therefore, if we combine the factory end with the automobile end and add some sensors and lidar at the factory end, it will help us find parking spaces faster and more effectively.
There is also a video, which is a cooperative system between Valeo and Cisco. In this process, pedestrians can be avoided, parking can be completed, and signals will be sent to our customers. When our users need to use the car, they can make an appointment in advance and greet our users from the automatic drop-off point. This is the concept of parking service. Valeo thinks that the combination of car terminal and factory terminal is a more effective and realistic scheme to complete parking service.
Another application scenario of automatic parking is very unexpected. What is this? Charge. You may not have thought about why charging is related to automatic parking at first. This is because, like autonomous driving, electrification is also a very big trend. You can see more and more plug-in hybrid cars and pure electric cars. These cars need to be recharged without exception, and plug-in hybrid vehicles may not need to be recharged so often.
Our survey of German users found that two-thirds of users think that if they can complete automatic charging or wireless charging, they are more willing to choose or use pure electric vehicles. I think it might be something. Because of the plug-in hybrid that has been on for almost two years, everyone found that charging guns are generally dirty and sometimes fall to the ground. When it rains, you don't want to pick up a wet charging gun. You'd rather have someone help you charge it automatically or wirelessly. A concept of Valeo, we can establish automatic and wireless charging through high-precision automatic parking, or use a robot to help you charge by wire. The range of this error, the accuracy must be improved to less than 10 cm. Even if charging, you should not think that charging can be completed near the charging pile or charging board. You need to have an accuracy. As long as the user stops the car once, he can automatically return to the parking position next time, and there is also an automatic avoidance.
This is automatic parking to complete charging, which requires high accuracy. Just said it was within 10 cm.
But let's think about it. Besides parking, what else do you need if you really drive automatically? In addition to the perception function, the most important point is positioning. Perception is just perceiving the surrounding environment, just like our eyes. But if you don't know where you are now, how can you really drive automatically? Generally speaking, for autonomous driving, the positioning method we can think of is to use GPS signals, but GPS, even in good weather, what our GPS can do is meter-level accuracy, which is about 2-3 meters. For navigation, GPS is no problem, as long as you know which road you are on. But the error of 2-3 meters is almost the width of a lane, which means you don't know which lane you are in. Navigation can't tell you whether you are on the side road or overhead. And if our lane is two-way and two-lane, it is very likely that the mistake of one lane will become retrograde, or you don't know that you are at the intersection when you are navigating at the intersection, so it is too late to tell you to turn. Therefore, for navigation, people may add their own perception, observe the surrounding environment and accept the accuracy of meters. But autopilot is unacceptable, and we need to improve it to centimeter level, which raises a big question, how to help autopilot achieve centimeter level accuracy, so we propose a RTK method here. At CES in 2020, Hyundai Motor, high-tech company Hexagon—Novatel, Valeo and mobile network operators all proposed high-precision joint positioning technology. Its significance is that after we use the GPS signal, you can get its high-precision position information with the ground base station and the ground base station in advance, and then make a differential comparison, and you can get a relatively high-precision position. This is the so-called RTK technology, which is a real-time dynamic differential positioning technology. This technology can help us achieve centimeter-level accuracy.
This is not a new technology. Hyundai Motor will carry this technology on his car in the future for quantification. This is already a high-precision technology that can be standardized for mass production.
RTK technology can help us achieve centimeter-level accuracy, which has been proved, but there are still limitations, such as what is needed for GPS signals? The weather is fine. If it rains today and the clouds are low, the GPS signal will be covered. What is the other situation? For example, if we go to a big city, such as Shanghai or Hong Kong, where there are many tall buildings, there is another restriction in Hong Kong, that is, there are many double-decker buses or sightseeing bus in Hong Kong, which will affect the signal, not to mention tunnels and viaducts, and the signal will definitely be affected. At this time, another technology is needed to make up or supplement the positioning, that is, we often use the point cloud technology of lidar to help the positioning. That is to say, we first build a high-precision map by lidar, and then compare the differences of high-precision maps in real time through the sensors in the car and lidar to help us locate relatively. This technology is actually very mature. We in Valeo use lidar to build such a high-precision map for real-time positioning. This high-precision map is in the form of crowdsourcing, because it is impossible to send many cars to update these maps in real time every time, so it helps to update this map in real time through the point cloud of our users' laser radar during use, so this is a form of crowdsourcing or crowdfunding. This method can supplement the RTK mentioned just now.
What's interesting? Under normal circumstances, in the case of compensation for tall buildings, the signal may be weak at that time because there is such a system to help locate through point clouds. On the contrary, when the GPS signal is unaffected and relatively empty, such as in the desert or desert area in the northwest, the geographical features are not so obvious. How do you locate it? At this time, RTK technology and GPS signals are used to make up for it. To some extent, these two technologies can complement each other and support each other, which can help us complete the high-precision positioning of autonomous driving.
At this year's CES, we also made a demonstration. Valeo is equipped with the second generation ScaLa lidar vehicle as a high-precision collection vehicle, and the first generation lidar fleet vehicle to show our high-precision vehicles, which will be displayed on the streets of Las Vegas in real time. In this case, we can find that our positioning accuracy can be improved to centimeter level, about below 10- 12 cm, which is a relatively high-precision positioning.
What needs to be told here is that ScaLa's first and second generation lidar are mass-produced lidar. At the same time, there is a roof positioning kit on the right. What does this mean? Generally speaking, laser radar and millimeter wave radar, like other sensors, usually cooperate with our OEM customers once mass production, which requires a long-term calibration and development work. These lidar or millimeter wave radars are not what you think. I'll buy a radar and plug it in. It's not that simple. This is a long-term development and calibration work. For some start-ups, especially self-driving start-ups, he may not be able to bear such time cost and development cost, so Valeo recently introduced a concept called universal sensor suite, which means that some sensors are still limited to lidar and ultrasonic sensors and made into a standard suite. In other words, its geometric dimensions, such as the roof kit just mentioned, are calibrated in advance. For users, especially self-employment, he needs to do a lot less work, and the time cost and development cost will be greatly reduced. Moreover, these sensors are mass-produced gauge sensors, so their quality, including the consistency just mentioned, will be guaranteed.
In Las Vegas, the roofs of these high-precision display vehicles are equipped with lidar, which is a practical and efficient solution.
What is the technical difficulty in truly realizing autonomous driving? Mr. Meng of Didi just said that there are many road users on the road, that is, traffic users who share the road with you. What their next intention is, it may or may not be known in advance, and you can't predict their next path. This is very difficult.
Let me give you an extreme example. We saw many electric cars on the road, especially these little brothers who delivered food. He was driving an electric car while talking on the phone. He doesn't even know whether to turn left or right or brake the next second. How did you know? This is the biggest challenge.
I remember two years ago, I went to a city in the south to visit an autonomous driving startup. They invited me to make a self-driving demonstration car in their car and show it on the road. When I was driving, the car suddenly braked. What is the reason? Because there is a person standing on the sidewalk in front, the vehicle sees a person on the sidewalk because of the conservative algorithm. I don't know what this person will do next, whether he will cross the road on the sidewalk or stay on the road, stop conservatively, and then change lanes to bypass the road in front of pedestrians.
Generally, drivers will pass a rough judgment when driving, passing at low speed or bypassing, which is a very big challenge for self-driving cars. How to predict others, not only pedestrians, but also traffic users such as bicycles, electric cars and scooters? At this year's CES show, Valeo launched a MOVEPREDICT. AI judges whether this person's attention is still focused on traffic movements through artificial intelligence machine learning. If not, we can adopt a more conservative approach. If his attention is still on the flow, the next reaction may be different.
Then you can judge his next step, predict his attempt or intention, whether he wants to cross the road or not, and his behavior must be judged by artificial intelligence. Of course, this is only a probability problem, and it can't be predicted 100%, but this is our next goal. If you can't predict, you can only use the most conservative algorithm and driving, which should be dissatisfied with the feelings of our users. In this case, autonomous driving will become a chicken rib, and you will drive more conservatively than people. In this case, autopilot can't really find the landing scene.
As I said just now, in fact, in many cases, the delivery of goods may be more practical than the needs of passengers. This is also the reason why we signed a strategic cooperation agreement with Meituan at CES 20 19 to jointly develop the last mile unmanned delivery technology, or the last mile unmanned logistics vehicle. This is the agreement we reached with the American delegation last year.
June 5438 +2020 10, at this year's CES exhibition, we launched an unmanned logistics vehicle jointly developed by Valeo and Meituan. Due to the limited space, we made a simple demonstration in a parking lot. There is a little brother in the picture. He doesn't have a remote control in his hand. Many people are asking whether he is controlling this vehicle like a remote control toy car. No, the only purpose is to start and end.
This is a year from signing a strategic cooperation agreement with Meituan, conducting technical exchanges, setting goals, and finally completing the design, manufacturing the prototype and transporting it to the United States. A lot of things have been done this year, and it is also a very fast process.
What kind of logistics vehicle is this? To give you a brief introduction, its size is 2.8 meters long and 1.2 meters wide, which is smaller than ordinary cars. You can send 17 takeout. This does not mean that it can only deliver 17. It has 17 delivery boxes, depending on the size of the takeaway, it may be able to carry more. The cruising range is electrically driven, totaling100km. If you need longer mileage, you need more batteries.
The division of labor between Valeo and Meituan is that Valeo provides such a drive-by-wire chassis, 48-volt battery system and controller, on which are the autopilot sensors and autopilot platforms provided by Valeo. The modules and software provided by Valeo are not only suitable for self-driving unmanned logistics vehicles, but also for vehicles under all urban road conditions. The car body provided by Meituan, including the car mentioned just now, as well as the distribution cabinet and APP, and the software exchange between users and customers is provided by Meituan.
This is a prototype car, which was made quickly in a year. Originally, our plan was to transport this car to Beijing for further display and exchange at the Beijing Auto Show in April this year. Because of the epidemic, this matter will definitely be postponed.
As I just introduced, in fact, the autonomous driving platform is an unmanned logistics vehicle. It is not specially built, but an automatic driving platform under urban road conditions launched by Valeo on 20 18 two years ago. This is the automatic driving aiming at L4 level under urban road conditions. It actually takes into account various characteristics of urban road conditions, such as various vehicles, pedestrians, bicycles, other traffic lights, including many roundabout in Europe, and stop signs. We also know the positioning of the vehicle through the high-precision positioning method just mentioned, so as to build an L4-level autopilot platform system.
We can take a look at this video, which is an autopilot demonstration made at the Paris Motor Show on 20 18. We need to remind everyone that all the sensors in this car are mass-produced by OEM and delivered to our end customers. Because it has been mass-produced, the driver is also using it.
This is a presentation made at the Paris Motor Show on 20 18. As you can see, a motorcycle has just passed by, and this is an automatic lane change overtaking. On the left is the camera in the car, on the right is the car behind the car, and in front is the scene of automatically avoiding the bicycle.
Traffic light recognition, zebra crossing, pedestrian recognition, avoidance, and finally tunnels and bridges can continue to maintain high-precision positioning under the coverage of GPS signals.
This is an automatic driving platform, and it is a system combining software and hardware.
If you look carefully, what is the sensor configuration of this unmanned logistics vehicle? It is equipped with various sensors, including four panoramic cameras, a long-distance forward-looking camera in front, four millimeter-wave radars, 12 ultrasonic sensors and four laser radars. The functions of the four lidar are different. The front and rear laser radars are used to detect obstacles, and the laser radars on both sides are more used to help high-precision positioning through point cloud maps. As you can see, there are four different sensors, and each sensor has a different amount of redundancy to complete a sensing function and help complete autonomous driving. These sensors have been mass-produced and we have used them in delivery.
But now many of them are relatively large, such as unmanned logistics vehicles, whose width exceeds 1 m and their length is 2 meters or even 3 meters. In fact, if you think about it carefully, it is difficult for these vehicles to enter the community and hotels. Because they are too big, they may have more contact with us, or use more small robots or small unmanned logistics vehicles. This is also at this year's CES show. We exhibited the unmanned delivery robot developed by Valeo in cooperation with the startup TwinswHeel. It may not be called a logistics car, but a robot. It has two wheels and four wheels. It's not autopilot, it's following you. For example, some old people or disabled people can't move when moving things. He needs a robot to help him carry goods or follow him. This is a scene. Valeo provides a 48 volt motor system with sensors. This startup has launched two unmanned delivery robots.
As long as you press this button, the sensor will know you. For example, if manager Zhou presses it there, it will know you. If someone presses it again, it won't talk to others. Just like dogs and pets.
This is another scene where unmanned logistics vehicles are used at home.
Valeo is the most complete sensor supplier. SCALA radar is the only one in the industry and the first one in mass production so far. The first generation of SCALA radar was mass-produced at 20 17. The third generation will be developed this year, which is solid-state lidar. The time is also determined according to our customers, probably around 2022.
In addition to OEM customers, there are also our start-ups or our self-driving companies. Here is an example of a French startup, which is equipped with Valeo's SCALA lidar. Valeo is also an investor in this company, accounting for about 10% of the shares. Since its establishment, the company has sold more than 65,438+060 self-driving cars in more than 20 countries around the world.
Finally, to sum up:
Like electrification or car sharing, autonomous driving is a very obvious and important trend in our "new four modernizations". Personally, I firmly believe that one day, we can really complete or realize driverless or autonomous driving. Of course, this road is long and may be bumpy, so I am a cautious optimist.
In this process, we should pay special attention to technology, but the more you develop into autonomous driving or highly autonomous driving, you will find that technology is only one of the problems. What else is there? Just now I talked about how to land, how to commercialize, and how to pay attention to the scene. I have repeatedly stressed that it is meaningless to talk about autonomous driving technology out of the scene, or it is hooliganism. We just talked about extreme examples. In an open place without any obstacles, any car can drive automatically at L4 and L5. But if you combine the scenes, you will find that many problems have appeared. What else do you need? Not only the automobile industry, but also our regulations, insurance, road construction and operators need to cooperate and work together to complete autonomous driving.
From this point of view, I am closer to Mr. Meng of Didi, that is, the possibility of self-driving of private cars, and it may take longer to land. Because I have already said that the cost of this autonomous driving must be borne by someone. I believe that every user here, you can't spend hundreds of thousands of dollars to buy a car and then spend hundreds of thousands to install an automatic driving system. It may be a faster, better and earlier taxi service provider, or it may be an unmanned minibus, an unmanned taxi or an unmanned logistics vehicle. We still can't tell which of the three landed first. However, unmanned logistics vehicles may pass the verification of this epidemic, and it may be easier to find some landing scenarios to complete the commercialization model.
In addition to these three scenarios, vehicles are also driven by L4 in mining areas and no man's land. In fact, a scene has been found, of course, this is relatively small.
But I want to sum up that autonomous driving is not just a private car, it definitely includes all kinds of vehicles in various scenes. I firmly believe that in this case, the scene of autonomous driving will not be far away, not ten or twenty years, but it may be faster, helping us achieve the goal of safer and more comfortable driving environment and logistics traffic.
Thank you for listening!
This article comes from car home, the author of the car manufacturer, and does not represent car home's position.