On June 22-23, 2020, the "2020 Third Global Automatic Driving Forum", jointly organized by Nanjing Economic and Technological Development Zone and Gaiji Automobile, was held. This forum focuses on core technologies, laws and regulations, technology evaluation, business models and other topics related to the commercialization of automated driving, and the following is the speech of Dr. Gu Jianmin, CTO of Valeo China, at the forum:
Valeo China CTO Gu Jianmin
Thank you for the invitation from Mr. Zhou, President of Gaixue Automobile, and I am very glad to have this opportunity to share with all the leaders, experts, and peers in the forum, and to share with them the latest developments in automated driving. Leaders, experts, peers to share, this topic is also very big, "From ADAS to the road of autonomous driving". I personally believe that active safety is an extension of the passive safety intelligence, if furthermore, ADAS is with the driver assistance system that we generally talk about, is an intelligent extension of active safety.
Isn't it true that autonomous driving is an intelligent extension of ADAS? In a sense it is, but autonomous driving is not just a matter of technology. As both speakers have mentioned this morning, it also involves scenarios, business models, and beyond that, there are regulations, infrastructure, and insurance that are all very relevant to automated driving, so what we're talking about today is not just a technology issue.
Because I'm speaking on behalf of Valeo, I believe that many of you know Valeo better, Valeo is headquartered in Paris, France, an automotive parts and components integration supplier, we are ranked in the top ten in the world, Valeo also has a lot of layouts in China, a **** has 35 factories, 12 research and development centers, there is a factory in Nanjing and a research and development center.
In terms of product lines, it can be said that if you are driving a car, there must be Valeo products or parts in this car. We have four divisions, one of the divisions of the main product is today we want to talk about the automatic driving driver assistance, in the product line inside the sensing system, is usually referred to as sensors, LiDAR, artificial intelligence advanced human-computer interaction, as well as the car network to provide to you, to help you create a meet the needs of everyone traveling, this is our company in the automatic driving a brief introduction.
If we look again, what is our topic today? From ADAS to autonomous driving, so I'm going to get right to the point and put this page here, this page of PPT, in fact, I've used it last year, and to this day I'm almost word for word, because the point hasn't changed.
The first sentence, how to do automatic driving, how to help the commercialization of automatic driving to the ground, what is the first thing? What's the best way to enter a market? Start small, that is, start with simple, low-cost autopilot technology. It's about technology here, starting with simple, low-cost technology.
What's next? The aim is to be used to attract enough users who are willing to pay, because as we all know, if autopilot is what? Demonstrate, test, no problem, everyone will welcome, but you can not yet commercial landing, what is the basic condition of commercial landing? Need someone to pay, the sky will not fall pie, always need someone to pay, either you pay, or our OEMs to pay.
How to do it? I listed below a few scenes or commercial landing or way, first from the automatic parking or valet parking, because we know that parking is relatively low-speed, and the scene is more controllable, in a semi-closed parking lot or parking garage. There are also from the low-speed automatic driving start to do, here is listed as 40 kilometers per hour, in fact, this speed has been very high. Generally speaking, the vehicles on the highway may be more than 40 kilometers. At low speeds, what can be done first? It can be a relatively small challenge for the sensing system, decision-making system pressure.
This starts with simple technology.
What else? Start with specific scenarios and specific uses. There are many scenarios for autonomous driving, and it doesn't make sense to talk about autonomous driving without running away from the scenarios. To take an extreme example, if you are in a test site, 300 meters in diameter, there are no vehicles, no obstacles, not to mention L4, L5 can be done. But in a different scenario, very congested, L3 can't even do it.
The key is to remove the safety driver, we have a lot of today's show test vehicles, self-driving on the road must have a safety officer, which is our current rules, regulations constraints.
But think about it, if there is a safety officer, we usually talk about L4 vehicles, or on the basis of L3, this aspect if we are not able to break through, our technology is still in the L3 this technology level, in essence.
Of course, another point of discussion today, in fact, the real self-driving do not get entangled in the end is L2, L3 or L4, we see today is to see the scene, how to break through the commercialization of the landing, to find a commercialization model, which is the most important.
Lastly, there is the need for delivery, which may be more practical than transportation of passengers. Of course it is not that the cargo may be a little less concerned than the guests from the point of view of security, which is not the only reason. If you see what we've seen in the past few months, especially when the outbreak was more severe, what do we see in cities like Wuhan and Beijing? There are some unmanned logistics vehicles to deliver medical equipment and medical supplies, which can avoid human contact, especially to places where the outbreak is more serious. This is also we can see unmanned logistics vehicles in this case, it may be more in demand than the delivery of people to deliver the needs of the scene.
That's one reason.
I'm going to throw out these points here.
Next please allow me to spend a little time in conjunction with Valeo's products, to tell you in detail how we find the scene to land, to find the commercialization of the landing of an ultimate goal.
Just now I said, automatic parking is a relatively easy to realize the scene, usually talking about automatic parking, parking assistance is what? Drivers need to be in the car according to the system's prompts, to complete the automatic parking or parking assistance. But once the driver moved to the car, so that our customers can choose to park in the car or outside the car, then it is remote control parking.
Valeo introduced remote parking as a feature in 2016, and it's also already in mass production, as you can see. With the remote key, in case of some emergency and need to stop parking, you can stop with one button.
Next can be a step closer, we can imagine, if we are in the underground garage mouth, we can use the remote control parking to let the vehicle automatically parked, with the remote control parking technology is much the same, but the difference is that one may be the vehicle needs to travel the distance or looking for a parking space in the scope of the larger; the second difference is that we are talking about valet parking, the need for factory support! The second difference is that we are talking about valet parking, which requires support from the factory. From the industry, there are two trends or two methods, one valet parking is completely rely on the vehicle side of the sensor to complete, another is the need for the factory side and the vehicle side of the coordinated completion of the valet parking.
If you rely on a sensor on the vehicle side, in a very crowded underground garage, it may take a long time to find a parking space, and at the same time may cause parking congestion. So if you combine the factory end with the vehicle end, you can add some sensors and LIDAR to the factory end to help us find parking spaces faster and more efficiently.
Here is also a video, this is Valeo and Cisco cooperation of a system, in the process, can avoid pedestrians, complete the parking, will send a signal to our customers, wait until our users need to use the car can be booked in advance, from the automated to the drop-off point to meet our users, this is the concept of valet parking. Valeo believes that the combination of car and factory is a more effective and realistic solution to accomplish valet parking.
Another application scenario for automated parking, quite unexpectedly, is what? Charging. We may not think at first why charging and automatic parking related? This is because at present, like autonomous driving, electrification is also a very big trend, we can see more and more plug-in hybrid vehicles and pure electric vehicles, these vehicles all need to be charged, perhaps plug-in hybrid charging does not need to be so frequent.
Our research with German customers found that two-thirds of them would prefer to choose or use a purely electric vehicle if they could do automatic or wireless charging. What do I think the reason might be? Because I've been driving a plug-in hybrid for almost two years now, and I've found that the charging gun is usually dirty, sometimes it falls on the ground, and when it rains you don't want to pick up the wet charging gun, you'd rather have someone help you with automatic charging or wireless charging. Valeo's concept, we can build through high-precision auto-parking to complete the automatic, wireless charging, or with a robot to help you wire charging. This range of error, the precision must be improved to within 10 centimeters, even if it is charging, we do not think that to the charging pile or charging board near the charging can be completed charging, there needs to be a precision. As long as the user completes a parking, the next time you can automatically return to this parking position, and here there is an automatic avoidance.
This is the automatic parking to complete the charging, need a relatively high precision, just said to within 10 centimeters.
But let's think about what's needed for real autonomous driving besides parking. In addition to the perception function, there is one of the most important is the localization. Sensing is just sensing the surrounding environment, just like our eyes. But if you don't know where you are now, how can you really do automatic driving? Generally speaking, for autonomous driving, the method we can think of for positioning is to use GPS signals, but GPS, even in good weather conditions, our GPS can achieve meter-level accuracy, almost in the 2-3 meters of error. For navigation, there is no problem with GPS, you just need to know which road you are on. But a 2-3 meter error is almost the width of a lane, which means you don't know exactly which lane you are in. The navigation can't tell you when you're on an auxiliary lane or on top of an elevation. And with our lane lines, if we have two lanes in each direction, it's very possible that the error in one lane line could turn you into a retrograde lane, or if you're at an intersection and the navigator doesn't know you're at the intersection, and by the time he tells you, it's too late to make a turn. So for navigation, maybe a human plus their own perception, observing the environment around them, can still accept meter-level accuracy. But autonomous driving can't accept that, we need to go up to centimeter level, and that raises a big question, how to help autonomous driving reach centimeter level accuracy, so we're proposing another RTK approach here, and at CES 2020, Hyundai, along with Hexagon-Novatel, which is a high-tech company, and Valeo and mobile network operators, proposed a high-precision joint positioning technology, which means that after we use the GPS signal, but through the ground base station, the ground base station you can get its high-precision position information in advance, and then carry out a differential comparison, you can arrive at a relative position with high accuracy, this is the so-called RTK technology, real-time dynamic differential positioning technology, this technology can help us help us to achieve a high level of accuracy. This is the so-called RTK technology, which is real-time dynamic differential positioning technology, and this technology can help us achieve centimeter-level accuracy.
This is not a new technology, Hyundai will carry this technology on his car in the future to quantify, this is already a standardized mass production of high-precision technology.
RTK technology can help us achieve centimeter-level accuracy, which has been proven, but there are still limitations, such as GPS signal needs what? The weather is better, if it is raining like today and the clouds are low, the GPS signal is obscured. What is another situation? For example, we go to the big city, like Shanghai or Hong Kong, such as high-rise city, Hong Kong has another limitation, that is, Hong Kong has a lot of double-decker buses or sightseeing buses will affect the masking signal, not to mention through tunnels, viaducts, the signal will certainly be affected, at this time we need to make up for or supplement the positioning of another technology, that is, we often talk about the use of LiDAR point cloud technology to help localization. That is to say, we use LiDAR to create a high-precision map, and then through the vehicle's sensors, LiDAR to real-time comparison of high-precision map differences, to help us relative positioning, this technology is actually very mature. We Valeo is through a laser radar to create such a high-precision map, to real-time positioning. This high-precision map is through the form of crowdsourcing, because it is not possible to send a lot of cars to update these maps in real time every time, so it is through our users in the process of using his LiDAR point cloud, to help update the map in real time, so this is a kind of crowdsourcing or crowdfunding form. This is a crowdsourcing or crowdfunding approach that complements the RTK that I just mentioned.
What is very interesting? In general, in the case of high buildings to make up for the situation, because there is such a system to help locate through the point cloud, the signal may be weaker at that time. On the contrary, in the GPS signal is not affected, more open, such as in the Great Northwest is a desert or desert area, geographical features are not so obvious, how do you carry out the positioning? This time to use RTK technology, GPS signal to compensate. These two technologies can compensate for each other to some extent, supporting each other, can help us to complete the high-precision positioning of autonomous driving.
In this year's CES we also made a display, Valeo equipped with the second generation of ScaLa LiDAR vehicles, as a high-precision acquisition of vehicles, as well as the first generation of LiDAR fleet vehicles to show our high-precision vehicles, this real-time demonstration in the Las Vegas Strip above. In this case, we can find that our positioning accuracy can be improved to the centimeter level, probably under 10-12 centimeters, which is a relatively high precision positioning.
Here we need to tell you that ScaLa's first and second-generation LiDAR are both mass-produced LiDARs. Meanwhile on the right above this picture there is a roof positioning kit, what does it mean? Generally speaking, LiDAR and millimeter wave radar and other sensors, once to mass production, usually with our OEM customers, need to go through a long period of calibration and development work, these LiDAR or millimeter wave radar is not like what you think, I buy a radar plugged in, plug-and-play, it's not so simple, it's a long-term development work, calibration work. For some startups, especially self-driving startups, he may not be able to withstand such time and development costs, so Valeo has recently launched a so-called universal sensor kit concept, that is to say, we take some of the sensors, which are currently limited to LIDAR and ultrasonic sensors, and make it into a standard kit. That means that the geometry, such as the roof kit, has already been calibrated in advance, so for the user, especially for self-driving startups, he needs to do much less work, and the time and development costs will be greatly reduced. And these are already mass-produced automotive-grade sensors, so their quality, including the consistency I just mentioned, is guaranteed.
We have these high-precision positioning demonstration vehicles in Las Vegas that use LIDAR kits on the roof, which is a more practical and efficient solution.
What is one of the more discussed technical difficulties in realizing autonomous driving? Just now the drop's Mr. Meng also said, on the road there are a lot of road users, is to share the road with you traffic users, what is their next intention, it is also very likely or that there is no possibility to know in advance, you can't predict their next path, it's very difficult.
I'll give you an extreme example, we see a lot of electric vehicles on the road, especially these delivery boys, he was talking on the phone while driving his electric vehicle, he doesn't know whether he's going to go to the left or the right or brake the next second, how do you know? That's one of the biggest challenges.
I remember two years ago, I went to a city in the south to visit a self-driving startup, and they invited me to do a self-driving demonstration vehicle in their car, and to do a self-driving demonstration on the road, and then suddenly the vehicle braked while driving, and what was the reason? Because there was a guy standing on the sidewalk in front of it, and that vehicle, because of a more conservative algorithm, it saw a person on the sidewalk, and it didn't know what the person would do next, whether they would walk up the sidewalk and cross the road, or continue to stay on the road, and conservatively stop, and then change lanes, and go around the road in front of the pedestrian.
The average driver will drive through a rough judgment, pass at a low speed or go around the side, which is a very big challenge for self-driving vehicles. How do we predict what other, not just pedestrians, but cyclists, electric cars, scooters, these transportation users are going to do. We are at CES this year, Valeo launched another MOVEPREDICT.AI, through artificial intelligence machine learning methods, to determine whether the person's attention is still focused on the traffic action, and if not, we can through a more conservative approach, if his attention is still on the traffic, the next step of the reaction may be different.
Then you can also judge his next, predict his attempts or intentions, whether he is going to cross the street, and his movements are judged by AI. Of course, this is only a probability problem, and not be able to predict 100%, but this is our next goal, if you can not predict, you can only use the most conservative algorithms and driving, which should be unsatisfactory to the feelings of our users, in this case, auto-pilot into a chicken ribs, you drive more conservative than the people are still slow, in this case, the auto-pilot and can not really find the landing of the scene.
In the earlier mentioned, in fact, in many cases the transportation of goods may be more practical than the demand for transportation of passengers, which is why we signed a strategic cooperation agreement with Meituan above the CES 2019, *** with the development of the last kilometer of unmanned delivery technology, or called the last kilometer of the unmanned logistics vehicle. This is an agreement we made with Meituan last year.
In January 2020, at this year's CES, we unveiled the unmanned logistics vehicle developed by Valeo and Meituan***. Because of site constraints, we did a simple demonstration of circling inside a parking lot. In the picture, there is a guy who is not holding a remote control, and many people are asking if he is controlling the vehicle like a remote control toy car. No, the only purpose is to start and finish.
This is what we did in one year, from signing a strategic cooperation agreement with the US group, technical exchanges, setting goals, and finally completing the design and manufacturing of the prototype vehicle, which was shipped to the United States. A lot was done in that one year, and it was a very fast process.
What kind of logistics vehicle is this? Briefly tell you about it, its dimensions are 2.8 meters long and 1.2 meters wide, which is a little smaller than the average small car. It can deliver 17 takeaways, this is not to say 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 range is electric, 100 kilometers a ****, with more batteries if you need longer range.
The division of labor between Valeo and MMT is that Valeo provides such a wire-controlled chassis, a 48-volt battery system, a controller, and on top of that Valeo's self-driving sensors, self-driving platforms, and modules and software that Valeo provides, not only for the self-driving unmanned logistics vehicles, but also for the self-driving modules for all the vehicles that are used in urban road conditions. Meituan provides the body of the vehicle, including the carriages mentioned earlier, as well as the delivery cabinet and APP, the software communication between the user and the customer, which is provided by Meituan.
This is a prototype car, within a year quickly made, originally our plan is in April of this year's Beijing Auto Show, the car to Beijing to do further demonstrations and exchanges, because of the epidemic, this thing will certainly be postponed.
I just introduced, in fact, the automatic driving platform for unmanned logistics vehicle, it is not specially built, is Valeo in two years ago, 2018 has launched a city road working conditions of automatic driving platform. This is in urban road conditions targeting L4 level autonomous driving, it actually takes into account the various characteristics under urban road conditions, such as a variety of vehicles, pedestrians, bicycles, other traffic lights, including in Europe there are a lot of traffic circles, as well as stop signs, all of these are taken into account. We also know the positioning of the vehicle through the high-precision positioning method mentioned earlier to create an L4-level autonomous driving platform system.
We can take a look at this video, this is a demonstration of autonomous driving done in 2018 at the Paris Motor Show above, need to remind you a little bit, in this car above all the sensors are all have been mass-produced, all have been delivered to our end-customers, in the OEM. because under the driver's hand is already mass-produced, in the application of the sensors.
This is a demonstration made at the 2018 Paris Motor Show, you can see that just now is a motorcycle passing by, the following is an automatic lane change, overtaking, the left is a camera inside the car, the right is the back of the car to follow the car to shoot, in front of the front is automatically avoiding the bike a scene.
Traffic light recognition, crosswalk, pedestrian recognition, avoidance, and finally tunnels, bridges, GPS signals are covered, can continue to maintain high precision positioning.
This is an autonomous driving platform, a system that combines software and hardware.
If we look in detail, in this unmanned logistics vehicle sensor configuration is how? Equipped with a variety of sensors, first of all, there are four surround-view cameras, there is a long-distance front view camera, four millimeter wave radar, 12 ultrasonic sensors, four LiDAR, four LiDAR's role is also a little bit different, before and after the LiDAR is to play the role of detecting obstacles, the two sides of the LiDAR's greater role is to be used to help high-precision positioning through the point cloud maps. We can see that there are four different types of sensors equipped, each sensor has a variety of different numbers, to complete a redundancy of sensory functions, to help complete the automatic driving. All of these sensors are already in mass production, and we're delivering them to customers who are already using them.
Just now said a lot of is relatively large, such as more than 1 meter wide, 2 meters long or even 3 meters of unmanned logistics vehicles. In fact, if we think about it, and finally enter the community, enter the hotel, these vehicles are difficult to enter, because too large, so may be more in contact with us more or use more is some small robots or small unmanned logistics vehicles. This is also in this year's CES show above, we showed Valeo and a startup TwinswHeel developed unmanned delivery robots, there may not be called logistics vehicles, called robots, there are two wheels, there are four wheels, it is not self-driving, it is to follow you, such as some of the mobility of elderly people or disabled people, he can not move things when moving, he He needs a robot to help him carry the goods or follow him around. It's a scenario where Valeo provides the sensor's 48-volt motor system, and the startup has now launched two kinds of unmanned delivery robots.
As soon as you press this button, the sensor recognizes you, for example, when Mr. Zhou presses it there, it recognizes you, and if someone else presses it again, it won't follow someone else. It's like a dog, a pet.
This is another scenario of unmanned logistics vehicles in home use.
Valeo is the supplier that has launched the most complete variety or types of sensors, and SCALA radar is the only one in the industry up to today and the first one that has been mass-produced to meet the automotive-grade LIDAR, and 2017 was the first generation of SCALA radar mass-produced, and this year, we'll be working on the third generation, which is a solid-state LIDAR, and the timing will also be based on our customers, which could be around 2022.
Laser radar, in addition to the general OEM customers, there are also our startups or our automated driving enterprises, here is an example of a French startup, this enterprise is equipped with Valeo's SCALA laser radar, Valeo is also an investor in this enterprise, accounting for roughly more than ten percent of the shares, this enterprise from the founding to the Today, the company has sold more than 160 self-driving unmanned minibuses in more than 20 countries around the world, as well as driverless cabs.
In conclusion:
Autonomous driving, like electrification or ****-enabled vehicles, is a very obvious and important trend in the "new four". I personally believe that one day, we can really complete or do driverless or autonomous driving. Of course, this road is quite long and may be very rough, so I am a cautious optimist.
In this process, we have to pay special attention to the technology, but the more we get to autonomous driving or highly automated driving, you'll find that the technology is just one of the issues. What else? Just said how to land, how to commercialize the landing, how to focus on the scene? I have repeatedly emphasized that it is meaningless or hooliganism to talk about autonomous driving technology in isolation from the scenario, and we have just talked about extreme examples. In an open space, without any obstacles, any car can be L4, L5 autonomous driving. But if you combine the scenarios, you will find that many problems arise, what else is needed? Not only the automobile industry, but also need our regulations, insurance, road construction, the operator of all aspects to cooperate, work together to complete the automatic driving.
From this point of view, I am closer to the point of view of Mr. Meng of DDT, that is, the possibility of self-driving private cars, may be a little farther away from the point of time to land. Because I've already said that the cost of this self-driving has to be borne by someone. I believe that every one of our users here, you can not spend hundreds of thousands of dollars to buy a car, and then spend hundreds of thousands of dollars to add a set of automatic driving system. Faster and better earlier landing may be rental service providers, there may be unmanned minibuses, unmanned cabs or unmanned logistics vehicles and so on, which of these three first landing, we can not see clearly. But perhaps the unmanned logistics vehicle through the verification of this epidemic, it may be easier to find some landing scenes to complete the commercialization model.
In addition to these three scenarios, in the mining area, no man's land, etc., is also a kind of L4 driving vehicles, in fact, has found a kind of scene, of course, this is relatively small.
But I want to summarize one point, autonomous driving is not just private cars, it certainly includes a variety of vehicles under various scenarios. I strongly believe that in this case, the landing of autonomous driving scenarios will not be very far away, not in ten or twenty years, but probably much sooner, to help us accomplish the goal of a safer and more comfortable driving environment and logistics and transportation.
Thank you for listening!
This article comes from the authors of Automotive House Car Family, and does not represent the viewpoint position of Automotive House.