At the beginning of the year, Great Wall Motor announced its 2021 sales results: its annual sales exceeded 1.28 million units, an increase of 15.2% year-on-year. This is the 6th year that Great Wall Motors has achieved sales of more than one million, and it is also the best annual sales performance in history.
A noteworthy thing is that most of the current models on sale of Great Wall Motors have been pre-installed L2 level intelligent auxiliary driving, as a product line of many enterprises, whether it is Haval, Wei brand, or pure electric brand Euler, and even hardcore off-road tank series and Great Wall Motor Gun all to achieve vehicle networking application coverage. It is not difficult to see that the Great Wall does not intend to stand idly by for the wave of automobile intelligence.
In fact, Great Wall Motors has been laying out in the field of intelligent ecology for a long time, and 2021 is a frequent achievement in the field of intelligence: upgrading the "Coffee Intelligence 2.0", including a series of advanced technologies such as the new electrical and electronic architecture, the third-generation intelligent driving computing platform IDC 3.0, the NOH intelligent pilot assistance system, the self-developed cockpit system GC-OS and the intelligent wire control site.
In June 2021, Great Wall Motor announced that it would sell 4 million units worldwide by 2025, of which 80% would be new energy vehicles. If the transformation of energy form is the "main theme" of change, Great Wall Motor's "coffee intelligence" is undoubtedly the first movement of change.
From perceptual intelligence to cognitive intelligence
Tesla launched the "Assisted Driving by Navigation" function in 2019, which realizes that after setting the destination navigation on the vehicle map, the vehicle can automatically enter and exit the ramp, automatically overtake, and automatically change the speed with the road speed limit. From 2020 to 2021, Weilai, Xiaopeng and Ideal have followed up to launch this function. Since then, in everyone's mind, "guided driving according to navigation" has become the label of technology travel companies.
In November 2021, Great Wall Motors officially pushed the NOH intelligent pilot assisted driving function to Weipai Mocha users. From the perspective of user experience, this system can be competent for point-to-point pilot assistance driving ability in the scene covered by high-precision maps, which also breaks the pattern of the four exclusive navigation assisted driving functions of the global "Tri-Lite Weipeng".
A month later, Great Wall Motors demonstrated a demonstration video of the NoH assisted driving system in Weipai Mocha City. Over a 34-kilometer test, the test vehicle passed 24 intersections, 27 pedestrian crossings, 22 traffic lights, 5 unprotected crosswalks and two roundabouts. During the whole process, the test vehicle equipped with the city's NOH system traveled a total of 11 kilometers without manual takeover.
So, how did Great Wall Motors' intelligent driving achieve rapid research and development in a short period of time?
The level of perception determines the lower limit of intelligent driving and is also the basis for collecting data. The core sensing devices of Great Wall Motors' intelligent driving are lidar and cameras.
The cameras around the body are capable of achieving 360-degree coverage, and the data collected by the cameras is calculated through a Resnet network and then generated in two branches. One branch can track and extract targets for multilayered features, including the identification of lane lines, stop lines, curbs, vehicles, and traffic lights; the other branch is used for the generation of the system Free Space and scene recognition.
For point cloud maps generated by lidar, PointPillar's algorithm is used to greatly increase the speed of calculation. The system first reduces the dimensionality of the three-dimensional point cloud data to pseudo-two-dimensional, and then calculates it according to the general method.
The level of cognitive ability determines the upper limit of the actual performance of autonomous driving. Expressing driving behavior in a specific scenario has several influencing factors from the macro perspective of cognition: weather, road structure, traffic participants, traffic flow density, main vehicle route, collision risk and collision time distance. These attributes are mined and expressed from the existing data, and then clustered and classified, so that the system can accurately identify these special scenarios.
At the micro level of cognition, Great Wall Motors refines the control action of the car. Using the most basic start-stop action as an example, Great Wall Motors refines this process into 4 stages: steady-state follow-up, front car deceleration, front car brake stop, and front car start.
After the macroscopic scenes and microscopic actions are digitized, Great Wall Motors marks them in a targeted manner so that the models can be continuously trained in different scenarios. The vehicle execution layer uses the collected cases as a guide, and the simulation learning results from the specific actions of each car in different scenarios, that is, end-to-end simulation learning.
The vehicle-side platform and the cloud platform are two-pronged
With the increase in the number of sensors on the car side, the amount of data acquired by cameras, lidar and other sensing elements has increased sharply, which poses a severe challenge to the computing power of the on-board intelligent driving computing chip.
To this end, Great Wall Motors launched the intelligent driving computing platform - IDC 3.0. The platform uses a combination of Qualcomm SA8540P + SA9000P, and the single board computing power reaches 360T, and it can also be upgraded to 1440T through inter-board cascading. It is also the computing platform with the highest energy efficiency ratio with mass production capacity in the world.
As a large-power vehicle-side computing platform, IDC 3.0 can effectively support the on-board computing of the AI visual model and the screening, cleaning, desensitization and reflux of vehicle-side perception data, greatly improving the accuracy of data recognition.
IDC 3.0 uses the highest current PCIE 4.0 interface to effectively support the high-speed bidirectional return of data between CUP and NSP, reducing system latency and reaction time.
In addition, IDC 3.0 supports 6 Gigabit Ethernet with 6 Gbps inter-board data transfer capability and simultaneous access to up to 14 8-megapixel HD cameras, 8 high-resolution millimeter-wave radars, and 6 solid-state lidars. Such performance also leaves redundancy for the later implementation of Great Wall Motors' intelligent driving to higher-level L4\L5 and other full-scenario intelligent driving functions.
Although the performance of the computing platform at the end of the car is very strong, the small model of a car end is generally only responsible for a part of the perception task at best. Great Wall Motor has accumulated an actual road measurement mileage of 4 million kilometers by installing an intelligent driving system in front of the Weipai Moka, and the large amount of computing brought by such huge data can only be solved by relying on the cloud computing platform. At present, Great Wall Motors' own supercomputing center is also under construction, and the computing power of the cloud can be improved in the future.
Great Wall Motor realizes full-task perception through a large model based on Transformer Fundamental Model, which can find out the recognition errors that occur when the small model on the car end occurs in bad weather and missed detection and misexspection, correct it and then return to the car model for retraining, which greatly improves the accuracy of data capture and the speed of model training.
The "cognitive intelligence" of Great Wall Motors' smart cars is also reflected in the test method of data. AI models are iterated quickly, and the most effective way is to test their effectiveness with a simulation environment. The traditional simulation test is extremely inefficient, from the scene construction, setting the model to the simulation test time-consuming and laborious, often each person can only build 30 test environments per day.
Great Wall Motor has developed a set of automated transformation tools and parameter generalization tools for semantic scenes, which can automatically convert the description text of the scene library in CSS into simulation test scenarios, and can automatically generate more than 10,000 simulation test cases every day. And Great Wall Motors has accumulated hundreds of thousands of typical intelligent driving scenarios, most of which are Corner Cases, which can effectively verify the performance in complex scenarios of intelligent driving and quickly iterate.
Zhen Longbao, director of intelligent driving at Great Wall Motor, pointed out that the core competitiveness of driving has two points: AI iteration cost and AI iteration speed.
Gu Weihao, CEO of Zhixing, also believes that the reason why Tesla's assisted driving can rely only on visual solutions to achieve today's achievements is largely because they have achieved data closed loops, and more importantly, Tesla has 1 billion kilometers of actual mileage data.
The intelligent driving of Great Wall Motors wants to obtain the ability to quickly iterate, and the most important thing is to quickly pave the volume. Objectively speaking, Great Wall Motor's many million-level annual sales models have paved the way for large-scale mass production of intelligent driving.
Security is essential
If the technical solution is the main theme of intelligent driving, then safety is the basic disk of intelligent driving.
Great Wall Motors provides six redundancies for intelligent driving, including steering redundancy, brake redundancy, perception redundancy, control redundancy, architecture redundancy, and power supply redundancy.
Through the design of dual winding steering motor, dual power supply, dual sensing and dual communication, Great Wall Motors intelligent driving can ensure that the minimum steering power is 50% in all scenarios, and the design of ESP + IBooster realizes dual redundancy of braking.
In terms of perception, using a multi-source heterogeneous sensor solution, the light front perception integrates the design of 3 lidar, 3 mm Wave radar and 2 cameras, and the security is greatly improved compared with the traditional fusion method. In addition to the Qualcomm SA8540P + SA9000P main chip, the security redundant chip is Infineon TC397, ensuring that the controller can take over at any time in the event of an unexpected situation.
Through the configuration of dual power supplies, the intelligent driving system of Great Wall Motors can ensure that the main power supply functions normally within 5 minutes after failure, and the switching of the main and auxiliary power supplies only takes 500 microseconds, which is enough for the vehicle to drive into the safe area or the driver to take over. At present, the law stipulates that L3-level intelligent driving can still operate normally for 10 seconds after power failure, and the safety redundancy of this system is indeed more thoughtful.
The future of smart-by-wire chassis
On June 29, 2021, Great Wall Motors launched the world's first wire control chassis technology with full independent property rights, what imagination will this bring to intelligent driving?
The steer-by-wire chassis eliminates the need for a steering drive shaft between the traditional steering gear and steering tube column, allowing for an extended ride. At the same time, because there is no traditional hard connection, the direction mechanism is completely isolated from the vibration of the road surface, which is conducive to the control of NVH, and can simulate a richer steering feel.
And due to the decoupling of the line control chassis, in the future, through the cooperation of high-level intelligent driving, the central processing unit can coordinate and control the action decomposition of each system on its own, so as to realize the autonomous coordinated control of the whole vehicle.
Imagine that thanks to the control of steering, gear shifting, throttle brakes and suspension through electrical signals, in the future, it is even possible to retract the steering wheel and freely combine seats during high-level intelligent driving, so that the vehicle can achieve a real "second living space".
At the same time, Great Wall Motors has set up a safety backup of the steer-by-wire steering and line-controlled dynamic systems that are strongly related to driving safety, and the overall failure rate of the system is controlled at one part of a billion per hour. And in the extreme scenario of steering failure, the steering action can be completed by applying braking force to one-sided wheels. In the extreme scenario of brake failure, it is possible to reduce the speed of the vehicle by increasing the kinetic energy recovery force and generating a large drag force.
Write at the end
When it comes to intelligent driving, the first thing that comes to mind is often Tesla, Weilai, Xiaopeng and a number of new car-making companies, who will be immediately put in the spotlight when they blow in the field, while Great Wall Motors' investment in the field of intelligent driving is not proportional to the attention they receive.
What is not known to most people is that Great Wall Motors' self-driving technology has been developed since 2009. From 2014 to 2016, the Great Wall Motors team won the China Intelligent Vehicle Future Challenge for three consecutive years. At the Great Wall Motor Technology Festival in 2015, Great Wall Motors demonstrated its L3 level of autonomous driving, and the Haval H9 with ADAS function was also launched in the same year.
After more than 10 years of accumulation, Great Wall Motor's intelligent R&D team will also be expanded to 2,000 people in 2021. From 2019 to now, in just over two years, Great Wall Motor's self-developed intelligent driving products have been mass-produced at a staggering speed, and have accumulated 4 million kilometers of actual road test mileage.
At the time point of the intelligent transformation of automobiles, mastering the self-research ability means that they do not have to be "stuck in the neck" of the upstream supply chain, which also means that they have absolute right to speak in related fields. But self-research has never been effective at the beginning, before the formation of scale and commercialization effects, the road to technological iteration and improvement is a long way to go, and there are not many companies in traditional car companies that are really willing to invest a lot of money and put it into practice.
The approach of Great Wall Motors is also different from the L4/L5 autonomous driving research and development enterprises that have been delayed in landing, and all their research and development are served in mass production. In 2021, there will be 250,000 vehicles with effective data collection functions in all models of Great Wall Motors, and in this "data-driven" era, who has obtained more road test data means that the iteration speed of intelligent driving products will become faster and the actual performance will be more mature.
At present, Great Wall Motors has installed intelligent driving systems in five models such as Weipai, Tank and Haval, and in 2022, this figure will increase to 34 models, with an installed capacity of more than 300,000 vehicles, accounting for about 80% of the overall listed models. The annual progress of hundreds of thousands of people also gives Great Wall Motors an advantage in the iteration speed of intelligent driving.
People are always not stingy about their cheers and attention to new things, and the great changes in familiar characters are sometimes overlooked by us. When everyone focuses on the intelligent driving process of the new forces of car manufacturing, the intelligent research and development of Great Wall Motors is also quietly advancing.