Current location - Trademark Inquiry Complete Network - Futures platform - Exclusive
Exclusive
Chedongxi

Words? |? Han Xiao

Secret research and development for 6 years, the team size exceeds 2000 people! Huawei finally showed its trump card in the field of autonomous driving.

The "high-order" automatic driving scheme named ADS will soon board the production car in Q 1 in 2022, so that consumers can fully realize the automatic driving between "their own garage" and "company garage".

You are not mistaken, that is, the driving tasks of daily commuting are given to the car-this is the function of the L4 automatic driving system.

If it can be mass-produced, ADS will directly crush all existing L2 and L3 futures systems, which is a typical "high-dimensional and low-dimensional" way. I have to say that Huawei's shot this time is really weighty.

At the past Beijing Auto Show, the ADS scheme was briefly displayed in the form of PPT, but the key issues such as what functions are behind it, how to divide driving responsibilities and what technical details are not revealed.

Su Qing, President of Huawei Smart Car BU Intelligent Driving Product Line

After the National Day holiday, Che immediately had an exclusive conversation with Huawei smart car BU? Su Qing, president of ADS intelligent driving product line, can get a glimpse of the whole picture, technical details and R&D process of ADS in advance.

First, all commuting tasks are handed over to ADS? L4 is used as L2.

The full name of ADS is autonomy? Driving? The official Chinese name of the solution is "Huawei Advanced Autopilot System".

To understand advertising, we can start with the function and responsibility.

The first is function. ADS is to realize the continuous automatic driving experience on the whole route from the owner's "community garage" to the "company garage".

In other words, from the moment you get on the bus in your garage, the vehicle is responsible for all driving operations-driving out of the basement, driving on the city road, driving on the ring road, driving out of the ring road, driving into the city road and driving into the company basement.

Existing L2-level systems, including the futures L3 just released by Mercedes-Benz, are available in some areas, for example, in the sections with clear lanes (expressways), on roads without red lights such as lanes and intersections (Tesla FSD can wait for the red lights by itself), or in garages and other scenes.

ADS city automatic driving test video

At the same time, when the L2 system changes lanes, goes up and down ramps, or passes traffic lights (such as Tesla navigation? Open? Autopilot and FSD also need human drivers to give instructions and confirm road conditions, and the whole experience is not coherent.

ADS solves the above two problems. First, the full coverage of the above commuting scenarios has been achieved, and the ODD (operating area) far exceeds the existing L2 and futures L3. Second, changing lanes, getting on and off ramps and other traffic lights are all done by vehicles.

So functionally speaking, ADS is point-to-point L4 autopilot-which is why Huawei calls it a high-order autopilot scheme.

Of course, the technical architecture of ADS also comes from the L4 autopilot system, which will be discussed in detail below.

The second key point in understanding advertising is the division of responsibilities.

According to the standards of SAE or Ministry of Industry and Information Technology, Huawei ADS belongs to L2 automatic driving system-the system provides assistance, and the driving responsibility is borne by the driver.

ADS expressway automatic driving test video

Su Qing, president of Huawei's ADS intelligent driving product line, told the car that when the driver uses ADS, the vertical and horizontal control is completed by the vehicle, but the driver should always pay attention to the road conditions and be ready to take over.

"After installing the driver's attention monitoring system, ADS also allows the driver to let go, but the time allowed to let go is defined by the car company." Su Qing said.

Seeing this, it is actually easy to understand the advertisement-selling a passenger car with L4 autopilot to consumers, but the division of responsibilities is set according to L2 autopilot.

Let the driver become a safety officer, always monitor the operation of the system and control the vehicle at any time.

Second, challenge the rainy streets of Shanghai? Calmly Respond to China Traffic

Having said that, how did the advertisement perform? Through a road test video provided by Huawei, you can see the situation of the game.

Huawei advertising test video

The video was shot on the streets of Shanghai. It can be seen that there was moderate to heavy rain or even heavy rain that day. Rain sometimes even blurs the camera and easily affects sensors such as cameras and lidar. There are many social vehicles on the road, and the road conditions are very complicated.

When passing through a traffic light intersection, the vehicle's perception system recognizes the green light and decides to move on. There is no lane line on the ground at the intersection, but with the help of high-precision maps, the vehicles run smoothly according to the planned path.

The ADS test vehicle accurately identified eight traffic lights at the intersection.

Then the vehicle came to a complicated intersection and prepared to turn left.

The left viaduct will affect the positioning accuracy of vehicles, and there are eight traffic lights ahead, which is also a great challenge to the automatic driving system.

From the video screen, the ADS system accurately identified all the signal lights, and also knew that the straight green light would enter the left-turn waiting area. Then the green light came on and the test car turned left.

On the right side of a narrow lane full of other vehicles, an electric bicycle appears in the lane and the opposite lane, and both targets in the video are accurately identified, tracked and displayed in the generated 3D scene.

Identification of electric vehicles by ADS system

The vehicle continued to drive on the path, and suddenly a human tricycles appeared in front to cross the road. If not identified, there will be a collision risk.

The ADS system accurately identified this target, and the vehicle chose to slow down and then accelerate after the tricycle completed steering.

ADS test vehicle avoids human tricycles.

On the urban roads in China, there may be electric bicycles, human tricycles, even horse-drawn carriages and donkey carts, and at the same time, there will also be behaviors that do not obey traffic rules, such as retrograde, crossing the road at will, and congestion.

These status quo will make mature systems such as Tesla Autopilot unacceptable-it is impossible to identify multiple traffic participants, and the decision-making algorithm does not consider the behavior of not obeying traffic rules.

Then the test car drove around a roundabout, turned a small corner, and then merged into the traffic on the expressway.

Expressway is similar to expressway, although it looks simple-go straight, but the difficulty lies in changing lanes and overtaking and dealing with traffic jams.

When the vehicle is about to pass the Shanghai landmark nanpu bridge, it detects that the slow speed of the black car in front affects the driving. At the same time, after the left middle lane is free, take the initiative to change lanes to the left and keep driving in the middle lane, thus independently completing the lane-changing overtaking action.

When driving to the center of the bridge, a BMW 5 Series station wagon in the rear quickly overtook the test vehicle, and then merged into the front of the test vehicle at a shorter distance-this is the most feared incision of L2 autopilot system? In the scene.

ADS test vehicle avoids traffic jam.

L2 system is faced with this situation, or sudden braking will affect the driving experience. Or continue to drive indifferently, causing the driver to be forced to take over the vehicle urgently.

In the video screen, the ADS test vehicle decelerated from 60 km/h to 55 km/h, allowing the 5 Series Travel Edition to complete the line merging. When it goes far, it will return to the speed of 60 km/h.

This is a typical case that the L4-level autonomous driving technology architecture reduces the dimension of the L2-level system-both sensors and software algorithms are "crushing" levels.

Su Qing told Che that Huawei's ADS system is currently being tested in China, specifically for electric vehicles. In the course, scenes such as merging traffic and overtaking are optimized to provide China consumers with the most suitable autonomous driving experience.

Judging from this 4-minute road test video, the performance of Huawei ADS is quite good.

Third, bicycle intelligence realizes AVP? Team learning to broaden the use of scenarios

From the previous part, ADS is actually equivalent to using L4 level system as L2 level system. But ADS actually has a real L4 function-AVP automatic parking service.

At the two ends of the commuting scene, there are the garage of the user community and the garage of the company. The goal of ADS is to realize AVP function in these two parking lots.

"Users only need to drive the car into the warehouse once, and the system can learn the parking path and method of this parking lot." Su Qing said, "The next time you go to the garage, the parking process can be completed by the vehicle itself."

When parking manually for the first time, users can customize the place to get off-such as the entrance before entering the underground garage or somewhere behind the basement.

Su Qing emphasized that technically, AVP can allow drivers to get off and use it-an L4-level automatic driving process.

However, because China's laws have not stipulated whether the public parking lot can use the AVP function, Su Qing said, "Huawei does not recommend users to leave the vehicle to use the AVP function, but more to reduce the driving burden of users."

All functions of ADS, including AVP, are completely realized by bicycle intelligence, so theoretically all suitable parking lots can use AVP functions.

Coupled with the team learning function, users will be more and more convenient when using AVP.

AVP system can enjoy parking information. For example, if car A has been to parking lot B, car A will automatically build a 3D map of parking lot B, and send the 3D map, surrounding environment and other information back to the cloud, and then distribute it to other vehicles through OTA.

One day, when car C wants to park in parking lot B, it can use AVP function directly. With more and more vehicles using AVP function, more and more parking lots will be supported, and eventually it will become a "universal" function.

Fourth, how to realize advertising? Hit with L4 autopilot technology?

In Su Qing's view, there have been many L2 autopilot systems in the last two years, but most of the functions have strict application scope and restrictions, which can't be used in many road conditions and traffic conditions, and can't meet the commuting needs of consumers in China, mainly urban road conditions.

It is precisely because of this pain point that Huawei decided to develop an ADS system to directly solve the commuting problem. However, commuting involves many scenarios, especially urban road sections, and the complexity of the system increases exponentially.

"So ADS adopts L4 autonomous driving technology architecture," Su Qing said. "Otherwise, it won't be used at all."

ADS technical architecture

In terms of hardware configuration, the ADS scheme will use a hybrid solid-state lidar with two or three lines 100, as well as a dozen cameras and six millimeter-wave radars. It can be said that it is armed to the teeth, and the configuration does not lose L4 unmanned taxis at all.

The computing center is called ADCSC (autonomous? Driving? Central? Super? Computer) domain controller, with rich computing power.

In terms of software, Huawei adopts a variety of AI technologies in the sensing part, and directly fuses the point clouds generated by self-developed millimeter-wave radar and lidar, as well as the video images of the camera at the pixel level (that is, pre-fusion), thus ensuring the sensing ability.

Previously, some autonomous driving companies mostly used laser radar and camera for pre-fusion, while millimeter-wave radar directly fused the output target with the perception results of the first two.

Huawei has a self-developed millimeter-wave radar, which can get the original point cloud data of millimeter-wave radar, and at the same time, the three sensors are pre-fused at pixel level and structured data fusion, which is a further step in technology.

Some sensors used by ADS

ADS should realize urban automatic driving (L4 function, L2 responsibility division), that is to say, it should be able to handle various scenes such as traffic lights and intersections, and avoid various traffic participants such as pedestrians, bicycles, tricycles and takeaway brothers.

This means that the rule-based algorithm used in the decision-making part of the traditional L2 system is powerless, and AI technology needs to be introduced into the decision-making part.

Su Qing said that Huawei defined different cells in the decision-making part with rules as the framework, and then further introduced machine learning technology into each cell.

"Pure AI algorithm is uncontrollable. Only by combining the rule algorithm with AI technology can we balance the effect and security. " Su Qing said.

Of course, there is also a high-precision map system necessary for advanced automatic driving in the ADS scheme, which is also the key to realize full-time commuting automatic driving.

Fifth, the team learns to solve map and data problems.

In the face of complex physical world scenes, the automatic driving function is inseparable from high-precision maps, but using high-precision maps will bring two problems: areas without map data cannot use automatic driving, and high-precision map data is difficult to update in real time, thus affecting the automatic driving system.

In this regard, Su Shi said that Huawei's ADS autopilot team already has a set of solutions.

First of all, the full-segment commuting function of ADS will be opened to users city by city according to the coverage of the map. For example, give priority to first-tier cities, and then gradually cover second-and third-tier cities.

It is worth mentioning that Huawei has its own first-class surveying and mapping qualification and surveying and mapping team, and has its own high-precision surveying and mapping capabilities. At the same time, Huawei has also built a map platform, hoping to speed up the production of high-precision maps with other partners.

Secondly, under the relatively simple highway (expressway) and parking lot road conditions, the functions of ADS can be used without high-precision maps, such as automatic car following/overtaking, AVP and other functions.

According to Su Qing, this setting maximizes the ODD—— of ADS-commuter autopilot function is used when there is a high-precision map, and autopilot function can also be used where there is no map.

Thirdly, the team learning function can help update high-precision maps.

Vehicles equipped with ADS system have many sensors, and at least two high-beam laser radars and cameras can be used to collect road change data in daily driving.

When there are more and more ADS vehicles and more mileage, the update frequency of high-precision maps can be accelerated.

ADS can generate the surrounding map by itself.

"The base map production of high-precision maps still needs professional collection vehicles to complete, and ADS vehicles are only responsible for the collection and update of change data." Su Qing explained that ADS vehicles will also build their own road spectrum when driving. When the data of real-time perceived road condition, high-precision map and self-built road spectrum are inconsistent, the confidence of the three will be calculated to judge the vehicle behavior.

If the vehicle encounters an extreme situation that cannot be handled, it will first maintain a certain route and call the driver to take over.

In addition to enjoying AVP and map information, another key function of ADS's fleet learning function is to collect driving data, which can be used to train AI models in perception and decision-making systems, and ultimately improve system performance.

Tesla's autopilot system has a similar setting, called shadow mode.

In the past few years, Tesla has sold more than one million vehicles, and its autonomous driving system has traveled more than 3 billion miles (about 4.8 billion kilometers). The data collected by the motorcade continuously provides "nourishment" for the iteration of autonomous driving system, making autonomous driving the strongest L2 position today.

Su Qing told Che that the team learning mode of Huawei ADS will collect all kinds of data and send it back to the cloud. In addition to the road environment information mentioned above, when the driver takes over or has uncomfortable operations (such as sudden braking), the system will also send relevant data back to the cloud for improvement.

There are many sensors for self-driving vehicles, so it is not convenient for practical operation if too much data is returned. In order to solve this problem, the ADS system will preprocess the target data locally, simplify it into structured data, and then send it back.

6. The secret R&D team has more than 2,000 people in 6 years? Commercialization of the following year

Since 20 19, Huawei has revealed some work and layout in the automotive field. Previously, the outside world only knew that Huawei was developing an automatic driving system and did not know the technical details and gameplay.

Only one year later, the ADS scheme was presented to the public by the opportunity of the 2020 Beijing Auto Show. Moreover, when it was first published, it gave the heavy news of "L4 as L2" and "Automatic driving of all sections of commuter road", which was quite emotional.

"Regardless of the time of previous technology accumulation, only from the direct research and development of autonomous driving technology, the related research and development of ADS has a history of five or six years." Su Qing said with a smile.

According to him, Huawei began to secretly develop autonomous driving technology around 20 14, and the team size has rapidly expanded from the initial one or two hundred people to more than 2,000 people today.

Huawei automatic driving team Shanghai office area

Even on a global scale, this is one of the largest autonomous driving teams in the world. Only big groups like Baidu, Google Waymo, General Cruise and Uber have the ability to form such a team.

In terms of team composition, Su Qing said that the team of 2,000 people includes not only autopilot talents from the automobile industry, but also a large number of doctors who graduated from prestigious schools and related majors, as well as personnel from Huawei's mature product line.

For example, Su Qing is a veteran who has worked in Huawei for 20 years. He is the founder of Huawei Kirin chips and solutions and the co-founder of Hisilicon chips and solutions. 20 14 began to lead the research and development of ADS system.

The research and development of high-grade autonomous driving system relies heavily on actual road test data.

Google Waymo leads the world in technology. One of the key reasons is that it has the largest road test fleet in the world and accumulated the most road test data-more than 20 million kilometers.

Su Qing did not disclose how many road test miles Huawei has accumulated. He only said that at present, there are 400 or 500 cars in China (half of which are owned by car companies). The actual road test mileage is "the existence of top matching" in China.

At present, Baidu is the player with the longest road test mileage in domestic autonomous driving companies, which has reached 6 million kilometers. From this point of view, Huawei's self-driving road test mileage will not be less than 6 million kilometers.

In mass production, ADS has achieved good results.

Su Qing revealed that ADS has won the fixed points of many car companies. In the first quarter of 2022, a variety of models will be listed as ADS solutions, with cars and SUVs covered, mainly pure electric vehicles.

"There are many car brands here." Su Qing said confidently, "We are not futures. In the first quarter of 2022, users can buy a car directly, and they can use advertisements when buying. "

The ADS system uses at least two high-beam laser radars, multiple millimeter-wave radars and more than a dozen cameras. Does this luxury configuration of directly calling unmanned taxis make the scheme extremely expensive?

In this regard, Su Qing replied that the price of the ADS scheme belongs to the medium level, mainly targeting at more than 200,000 vehicles.

Now, the ordinary L2-class autopilot system is equipped on the model with a price of about 654.38+ million. The two L3-level futures systems that have been released need to be equipped with high-end automobile brands with a starting price of nearly one million.

In contrast, more than 200,000 models are equipped with the ADS system, which is also considered as a medium level.

Conclusion: Huawei provides a new idea for mass production of autonomous driving.

After years of development, the global autonomous driving industry is facing an embarrassing situation.

Advanced autonomous driving routes represented by Google Waymo, Baidu Apollo and other technology companies have made some breakthroughs, but it is still far from the large-scale deployment and technical realization of unmanned taxis.

On the gradual route represented by the traditional automobile factory, L2 automatic driving system has been rapidly popularized, but further L3 automatic driving has become a "technical black hole", so far no company has been able to achieve mass production.

In this context, the idea of "L4 as L2" used by Huawei ADS provides a very good way for the industry to "save the country by curve".

On the one hand, the function setting of "commuter autopilot" makes the autopilot system of passenger cars no longer a decoration, which can play a role in daily driving, and also helps car companies break through the ceiling of L2 autopilot.

On the other hand, since we can't solve all corners? Case ensures the absolute safety of the system, so the L4-level system should be used as the L2-level system in driving responsibility-the driver monitors the road conditions all the time, but the vertical and horizontal control is completed by the vehicle, which makes it possible for the automatic driving system to be mass-produced, while the L4-level automatic driving system is mass-produced in advance.

To some extent, the vehicle carrying ADS is an L4 unmanned vehicle, and the driver becomes a security officer.

More importantly, once mass production is achieved, hundreds of thousands and millions of motorcades will collect a large amount of data every day and send them back to Huawei, helping to finally break through extreme scenes and let human beings truly enter the driverless era where they can sleep while driving.

This article comes from car home, the author of the car manufacturer, and does not represent car home's position.