【EV Vision Technology Analysis】 "Technology changes life", this sentence is not false.
If the history of mankind is summarized in the theory of evolution, is it not true that the evolution of technology can also be written in it? Why say so, because in today's intelligent technology popularization, human beings living under big data can already make virtual intelligence a reality and integrate it into our lives, it can be said that this generation of human beings has used "moving their brains and mouths" to complete daily bits and pieces.
The same goes for driving.
Imagine that when we enter the car and are ready to drive, just a sentence of "where I want to go ~ which ~ where", the vehicle will automatically calculate the best route and automatically drive, and at this time sitting in the driver's seat we have changed from a driver's identity to a passenger, this is the advent of the automatic driving ecology.
However, if you want to achieve this realm, the functional level of automatic driving must be L3 level. Because this is a threshold for true autonomous driving capabilities, whoever can reach it can be called autonomous driving.
How to achieve true · L3 Autonomous Driving?
First, let's look at the diagram.
Last year, the Ministry of Industry and Information Technology officially released the recommended national standard of "Automobile Driving Automation Classification" for approval and publicity, and the formal standard is scheduled to be implemented nationwide on January 1, 2021. It is reported that the standard clarifies the grading principles, elements and determination methods, as well as the technical requirements of each grade. Based on the degree to which the driving automation system can perform dynamic driving tasks, according to the role assignment in the execution of dynamic driving tasks and whether there are design operating conditions, it is applicable to M-class and N-class cars, and other types of vehicles can be performed by reference. According to the standard, the total classification of driving automation is 6 levels, from low to high from 0 to 5.
Specific aspects: It can be seen that L0-L1 level is basically a normal driving state can not be mentioned, under the L2 level of automatic driving, when the automatic driving is turned on, your hand can leave the steering wheel for a short time, but the overall you are still in a driving state. The autopilot system is only responsible for helping you find the way, but the situation still requires your own intervention.
At L3 level, the driver can release both hands and feet at the same time, and the entire automatic driving system will control the vehicle according to the road conditions, but in the event of an emergency, the driver's intervention is required.
In addition to the way, in terms of hardware and software, L2 level and L3 level also have a distinct difference.
In L2-level autonomous driving systems, 9 to 19 sensors are required, including ultrasonic radar, millimeter wave radar, surround view cameras, and more.
In L3 autonomous driving, the size of this sensor will increase to 19 to 27, including lidar, high-precision maps, high-precision positioning and other software and hardware, and the cost will also increase.
Here to say, at that time, many car companies advertised that their models had L2.5 or L2+ level automatic driving capabilities, so was such a model close to L3 level automatic driving?
In fact, according to national standards or international standards, there is no L2.5 or L2+ scale, and their appearance is only to cater to the merchants to promote a new car. Because these companies are still some distance away from truly achieving L3-level autonomous driving, but the L2 level has been unable to attract more consumers, it will appear L2.5 or L2+ level, which is full of local characteristics.
Another point, in terms of safety, from the ISO 22179 standard (vehicle specification level) in the calculation of the braking distance, the general vehicle must at least perceive the target of about 200m in front of it to have enough early warning time, and on this premise, it is also necessary to consider the problem of redundancy, and finally on the issue of computing power, the L3 level of automatic driving functions require tens of billions of calculations per second, and it must not be delayed, otherwise there will be dangers leading to accidents.
It is worth mentioning here that the HD map plays a crucial role in the entire L3 level of automatic driving, and it also provides a guarantee for safety during driving. Because the electronic maps we use every day are based on commercial GPS, the accuracy is about 5m, and the positioning accuracy of the HIGH-precision map is 0.1m, which is because the high-precision map supplier will collect data on these lines in advance with a vehicle full of sensors before drawing, and calibrate and standardize these data, so any changes on the road surface will be prompted on the navigation map.
Baidu high-precision map collection car
However, at this stage, some models with L3 level automatic driving capabilities in the domestic market are restricted by some regulations, and cannot fully play the advantages of "taking off their hands and eyes", but in recent times, some individual cities and regions have begun to gradually relax restrictions.
Last year, the "Special Plan for the Intelligent Connected Vehicle Industry (2020-2025)" issued by Shanghai Municipality guided the promotion of highly automated driving (L3 level or above). Among them, it includes high-speed vehicles, elevated road tests, and pilots in specific areas to carry people without safety officers.
According to the plan, Lingang New Area should continue to give full play to the advantages of the free trade zone in institutional innovation, strive for the support of the relevant departments of the state and Shanghai Municipality, and declare the national-level vehicle networking pilot zone, the national quality inspection center, the transportation industry research and development center and the intelligent transportation demonstration project. It can be seen that Shanghai may become the first region in China to open L3 level autonomous driving on the road.
Who can reach the peak of L3 level in China?
According to the above paragraph, some domestic car companies have already had L3 level automatic driving capabilities in their models, including many new forces car manufacturers and independent old car companies, although they have their own characteristics in the system configuration, but highlight the overall performance of the country in this field.
Xiaopeng NGP
At present, the Xiaopeng P7 with XPILOT 3.0 system will obtain NGP autonomous driving functions through OTA upgrades, and its computing platform uses a dual combination of NVIDIA Xavier chip + intelligent controller. Xiaopeng P7 exterior includes 4 front cameras, 5 enhanced perception cameras, 4 surround view cameras; At the same time, there is also a camera in the car, a total of 14 cameras in the car, in addition to 5 millimeter wave radar, 12 ultrasonic sensors.
In terms of positioning capability, Xiaopeng P7 uses AutoNavi High Precision Map, which has an accuracy of decimeter-level, which is the basis for realizing lane-level positioning; and the use of triple-precision positioning hardware (GPS+RTK+IMU) makes the global positioning accuracy up to centimeter-level; in addition, the use of real-time positioning and map construction technology (SLAM) makes the P7 relative positioning accuracy less than 0.3%, and the triple technology complements and superimposes each other, which can effectively improve the stability and safety of automatic driving under overpasses, tunnels or bad weather.
GAC AEAN ADiGO 3.0
GAC Aeon's newly upgraded ADiGO 3.0 automatic driving system has the L3 level automatic driving capability of the full speed domain of the whole road section of the high-precision map, which can support the driver to control the vehicle for a long time on all highways and expressways in China when permitted by regulations. At present, the system has been successively deployed in the AION LX and AION V models.
In 2020, the AION LX equipped with ADiGO 3.0 autonomous driving was the first mass production and delivery, and due to the first application of China's high-precision map, the AION LX became the first mass-produced L3 autonomous driving model. Prior to this, GAC Aeon's ADiGO autonomous driving system (including 2.0) has been delivered to other models, and the ADiGO autonomous driving system has so far steadily driven more than 80 million kilometers.
AION LX's ADiGO 3.0 automatic driving system consists of 12 ultrasonic radars + 4 lateral millimeter wave radars + 1 forward millimeter wave radar + 4 panoramic cameras + 1 Mobileye EyeQ4 forward-looking camera, and integrates the world's only "high-precision map + high-precision radar + intelligent camera" triple perception technology, the detection distance reaches 200m, can achieve any angle recognition of the front car, obstacle recognition and multi-target tracking, can achieve 0-120km/ h Full speed domain free hands and feet driving.
As one of the cores of ADiGO 3.0 autonomous driving system, the hd map function is also used. The so-called high-precision map is a high-precision data map used to support vehicle automatic driving, which is the key to ensuring the safety of automatic driving, with centimeter-level accuracy and lane-level attribute information. The system's high-precision map is specially developed for the automatic driving system, managed by the intelligent driving decision controller, the accuracy can reach about 10cm, allowing the vehicle to accurately obtain lanes and related information, predict the road conditions of 1km ahead, and start decelerating 500m before entering the bend to ensure a safe speed through the curve (the camera-based perception system can only start decelerating 200m in advance at the earliest), so as to plan the driving route in advance.
NIO NOP
In September last year, NIO officially launched the Navigate on Pilot (NOP) function at the 2020 Beijing Auto Show. NOP is the second assisted driving system in the world after Tesla NOA that can autonomously complete up and down ramps, high-speed cruise, and lane changing overtaking based on navigation information. This function is based on adaptive cruise, lane departure suppression and other functions, through the access to the navigation system, can automatically change lanes to overtake, drive in/off the ramp, according to the road information to control the speed and other intelligent operations. In contrast to the "Driver Assistance System", automatic driver assistance at points A to B is realized.
In terms of sensors, NIO is equipped with a forward tri-eye camera installed under the windshield, a forward millimeter-wave radar mounted on the front bumper of the vehicle, and four angular millimeter-wave radars mounted on four corners, which are the main sensors that realize NOP and are used to provide environmental perception under automatic assisted driving. The driving state detection camera in the car is mainly used to detect the driver's state when the auxiliary driver is turned on to complete human-computer interaction. 12 ultrasonic radars and 4 surround view cameras are used for low-speed autonomous parking.
In terms of positioning, WEILAI Automobile uses Baidu's high-precision map, and is also the earliest batch of domestic automakers that cooperate with Baidu's high-precision maps, and the collection range of Baidu high-precision maps determines the scope of application of Weilai NOP, and the current collection range of Baidu high-precision maps is 300,000 kilometers of high-speed, national highways and urban expressways.
The vehicle through the camera perception of the number of lane lines, location, type, color and other information, and the lane of the HD map matching, determine which lane it is in, and then according to the curve equation of the lane line and the curve of the HD map to do fitting optimization, correct the lateral position of the self-vehicle in the lane, through the features on the ground (such as arrows) and the features on the map to match, correct the longitudinal position of the car, to achieve high-precision matching positioning.
Huawei ADS
The Polar Fox Alpha S Huawei HI Edition models are equipped with a full set of solutions for Huawei's smart car BU, and are also the first models to use Huawei's "HI" LOGO. It has an intelligent high-end automatic driving system equipped with 3 lidars, a Hongmeng OS intelligent interconnection cockpit, a 400TOPS hashrate chip, a 3.5-second zero-hundred acceleration, and a dual-redundant system, so it can reach Huawei's current highest level of automatic driving.
Specifically, ARCFOX Polar Fox Alpha S Huawei HI Edition is equipped with Huawei's most advanced three 96-wire car-grade lidars, with a maximum detection distance of 200 meters and a detection frame rate of 10 Hz. Equipped with 9 ADAS cameras, 12 ultrasonic radars, 6 millimeter wave radars, and 4 surround view cameras, it realizes super perception capabilities with 360° all-round coverage. In addition, the new car is also equipped with Huawei's self-developed chip with a computing power of 400Tops, and its matching high-precision map can accurately extract lane and traffic light information, and through continuous machine self-learning and continuous iteration, the control performance at the intersection is closer to manual driving, including functions such as straight forward, left turn, and unprotected turn.
EV Horizon Perspective:
To sum up, L0-L2 is a low-level driving system, while L3-L5 is an advanced automatic driving system. Among them, in the leap from L2 level to L3 level, the largest span is changed from the driver to the system by the monitoring body of the environment. High-level autonomous driving can only be possible if the system is able to automatically probe and analyze the condition of the nearby area. Therefore, to achieve true L3 automatic driving, high-computing chip + high-precision map + high-precision radar + intelligent camera are indispensable, of course, Tesla is an exception (camera + millimeter wave radar vision software solution). At present, the state allows the application of mass production to basically limit the automatic driving technology below L3 level, requiring users to actively control the car in real time while driving, and it is unsafe to transfer the driving right to the car to control. Therefore, manufacturers should also be absolutely cautious in publicity, do not exaggerate the promotion of products, and do not let users misunderstand that they can rest assured that the driving right will be given to the car! Therefore, the merchant must not let the user misread and lead to the occurrence of dangerous things in order to promote its own products, so that the merchant is to bear the corresponding consequences and responsibilities.
Although the policy has temporarily become a restrictive door for the automatic driving function on the road, it is gratifying that most domestic companies are still unremittingly researching and developing autonomous driving. Just before the Shanghai Auto Show, some car companies made successful demonstrations in autonomous driving. Among them, Didi's continuous 5-hour urban driverless function test; Huawei and Jihu Automobile's advanced autonomous driving road test under complex working conditions; WM's L4 functional test; Xiaopeng's automatic driving from Guangzhou to Beijing is nearly 3,000km, and the number of full takeovers is about 20 times, which also means that it is about 0.7 to 0.8 times per 100 kilometers. It can be seen that the development of autonomous driving technology in China has gradually led the world, which is a crucial progress for us to "overtake in curves" in developed countries.