【EV Vision Report】On December 23, 2021, the first "HAOMO AI DAY" after the upgrade of the brand day was officially staged. The HAOMO AI DAY set up venues in Beijing and Baoding, and Zhang Kai, chairman of Mo mo Chi Heng, and Gu Weihao, CEO of Mo Mo Zhi Xing, led the partners of Mo Mo to announce the latest report card and the latest release of Mo Mo Zhi Heng.
The end of the attack road
After the press conference just began, the first person to bring a speech to everyone was Zhang Kai, chairman of Zhixing. The theme of his speech was "The Road to Attack at the End of the Day". In the Hall of Baoding Haval, where he was located, the latest models of Great Wall Motors, wei brand mocha, tank 300 and other models, have been equipped with the assisted driving system of the millimeter.
Zhang Kai introduced that this quarter, the Weipai latte, Macchido and Haval Divine Beast equipped with the Millima Zhixing Auxiliary Driving System are also actively preparing for mass production and listing. This month, Zhixing will also usher in the mass production of the 1,000th terminal unmanned logistics vehicle off the production line.
In the past two years, The Wisdom of the Millima has worked hard and determined to move forward, ranking among the first echelon of China's automatic driving mass production capacity at the fastest speed, and creating the "Millide Mode" of China's automatic driving.
Facing the upcoming 2022, Zhang Kai first made a judgment: "2022 will be the most critical year for the development of the automatic driving industry, the competition in the field of passenger car assisted driving will officially enter the second half, and the automatic driving of other scenarios will also officially enter the first year of commercialization." Specifically, he brings ten predictions for industry trends.
Trend 1:
In terms of the development of autonomous driving technology, Zhang Kai believes that data intelligence will become the positive or negative hand of automatic driving mass production. The data intelligence system is the key to the closed loop of automatic driving commercialization, and building an efficient and low-cost data intelligence system is the foundation for the healthy development of automatic driving, and it is also an important link for the automatic driving system to continuously iterate forward.
Trend 2:
Transformer and CNN are deeply integrated and will become the glue for the integration of autonomous driving algorithms. Transformer technology can help the autonomous driving perception system to understand the environmental semantics more deeply, and the deep integration with CNN technology will solve the problem of mass production and deployment of AI large models, and will become a key technology in the second half of the competition in the autonomous driving industry.
Trend three:
With the development of AI technology, the large computing platform will be officially mass-produced in 2022, and Transformer technology and ONESTAGE CNN technology need to be supported by a large computing platform. At present, the market head players are laying out large computing platforms, in fact, they are not simple material stacking, but the needs of the development of a new generation of AI technology and algorithms. The computing platform and the new generation of AI technology it carries will push autonomous driving technology to a new stage.
Trend 4:
In terms of mass production landing, 2022 will be the first year of mass production of urban smart pilot auxiliary driving. The competition for passenger car assisted driving systems will officially enter the second half, and the scene of competition in the second half will be the urban opening scene. The launch of the Urban Pilot Assisted Driving System will push the experience of intelligent cars to a new height.
Trend Five:
The commercialization of unmanned logistics at the end will show continuity. This year, in the field of unmanned logistics distribution at the end, only the last one has achieved mass production and sales of 1,000 vehicles. In 2022, we hope to take advantage of the advantages of millimeters to expand the market capacity in this field by 3 times. Therefore, I hope that this field can achieve commercial closed loop in 2022, and at the same time, the leading players in this field will begin to try the feasibility of scene replication.
Trend Six:
Robotruck will officially start the exploration of the road to mass production. Compared to the complexity of the Robotaxi running scenario, Robotruck runs on the highway for a long time, and the scene is relatively simple. Robotruck will take a gradual development route from assisted driving to unmanned driving, in terms of the market's willingness to pay, Robotruck can effectively alleviate driving fatigue, reduce operating costs, and have the feasibility of commercialization of closed loops. However, as far as the current experience of Zhixing is concerned, the mass production of autonomous driving systems will be a hurdle, so Zhang Kai believes that the exploration of robotruck mass production will be the only way.
Trend Seven:
In terms of industry development trends, the smart car market based on intelligent driving technology will converge on the head of the automotive industry, which is similar to the smartphone market in previous years. The initial sales volume of the smart phone market showed an inverted triangle market form, with the advent of the era of fierce market competition, today's smart phone market sales show a T-shaped market form, and the head effect is very obvious. Zhang Kai said that 2022 will be the beginning of the era of full competition for smart cars, and intelligent driving technology will accelerate the arrival of the era of full competition for smart cars.
Trend Eight:
With the mass production and scale of the automatic driving system, the AI perception technology composed of lidar and machine vision will achieve deep integration of lidar applications with the large computing power computing platform, which greatly enhances the accuracy and certainty of the automatic driving perception ability. The deep integration of AI perception technology composed of lidar and machine vision and the computing platform with large computing power will greatly improve the operation efficiency of automatic driving perception and cognition modules.
Trend Nine:
Safety will still become the top priority of the automatic driving system, functional safety, information security, and expected use safety are fully considered in the design of the automatic driving system in the early stage, and now there is more data security and data compliance. The Automotive Data Security Law has been officially implemented in October this year, and it is expected that in 2022, the national level will issue detailed rules to enforce the data security law, which will further increase the closed-loop difficulty and cost of the automatic driving data intelligence system.
Trend 10:
2022 will be the most intense year of competition for AI autonomous driving talents. At present, 70% of the scientific and technological innovation in the automotive industry comes from the field of autonomous driving technology, and Europe, the United States, China and Japan are continuing to invest in this field on a large scale. Behind the large-scale investment is the competition for AI autonomous driving talents, China has been at the forefront of the world in the field of automotive intelligence, Chinese consumers' acceptance and expectation of autonomous driving, intelligent cars are significantly higher than developed countries in Europe and the United States, it can be seen that in 2022, China will become the most fierce competition for AI autonomous driving talents.
In 2021, through the unremitting efforts of the team, the smooth mass production of five models was achieved, becoming the most prolific company among domestic autonomous driving companies. In November this year, Miller Zhixing successfully realized the mass production and listing of the intelligent pilot auxiliary driving system NOH, which was approved in November 2020, completed the system design matching and debugging in four months, and completed the engineering development of the system in four months, and the road test verification and system polishing in parallel for eight months, making the terminal zhixing become the fastest AI autonomous driving company in the development and landing of the domestic NOH intelligent pilot auxiliary driving system.
The road to AI at the end of the day
After Zhang Kai, chairman of Miller Wisdom Bank, finished his speech, Gu Weihao, CEO of Miller Wisdom Bank, delivered a speech with the theme of "The Road to AI at Millima". He mentioned that friends who have known Mo Mo early will understand that Mo Mo's windmill strategy, a data intelligence center, and three business directions, including passenger cars, unmanned logistics vehicles, and intelligent hardware. In the passenger car direction, this quarter added three models, Macchiato, latte, Haval Mythical Beast, so far, in the 2 years since the establishment of the milli- end of the 2 years, the number of passenger car models equipped with milli-terminal products has reached 5, 5 unmanned logistics vehicle models, and 2 unmanned follow-up equipment. In 24 months, We have created 12 products that create tremendous value for our customers.
The first product of Zhixing in passenger cars, Shiwei brand mocha, began gradual mass production after this year's auto show, in these 6 or 7 months, users are very fond of the driving assistance function we provide, so far, the small magic box auxiliary driving mileage has exceeded 4 million kilometers.
In the use of these 4 million kilometers, what did Zhixing find?
Zhixing found a large number of situations that could not be imagined before mass production. Discover that the real world is far more complex than we thought. Different weather, roads, traffic participants, traffic flow density, and conventional driving habits constitute a rich real world.
In many scenarios, we humans driving will be very challenging. Pedestrians who need to be careful out of the blind spot need to participate in each other's game with traffic, and they also need to face unruly traffic participants. In the use of our current system users, we found that the three main user takeover scenarios: the sense of oppression brought by the passage of large vehicles, the irregular sections such as lane lines brought about by road construction, and the irregular entry of surrounding vehicles, each scenario is worthy of our careful response and challenge.
In this evolutionary process, Zhixing summarized the curve of autonomous driving capability development, which is a function related to the scale of data, F=Z+M(X). Where F stands for the product power of the product, Z represents the first generation of YY products that we sit in the office, and M is a function that turns data into knowledge. Including: data acquisition, data expression, data storage, data transmission, data calculation, data validation, all of these considerations, plus the impact on cost and speed.
These factors, if the product is not mass production, if the mass production is less than tens of thousands of scale, is difficult to realize the problem. Starting from autonomous driving, the existing data structure of human beings will have huge structural changes. Now in the storage content of human beings, the value of text data dominates, and many photos have been added to the mobile Internet era, and the proportion of image data will become larger and larger in the future. This will affect the development of many industries, and I will talk about some of them later.
So this M, facing the data is image-based data, facing the problem, is a lot of new problems, there will be a lot of new challenges. Acquisition, including the record of data, data selection and data compression expression, including the definition of data, data associated storage, is easier to understand, but when the amount of data increases rapidly, under the current storage mechanism, it will face different cost-effective storage solutions.
Transmission includes on-chip data transmission, including data transmission between heterogeneous chips within the same hardware, and data transmission from the end to the cloud. Computation, here there will be the concept of our algorithm, there will be all the AI algorithms in order to complete a specific task of the calculation. Verification, including comparative verification of calculation results and simulation verification. M is complex, the end of the wisdom in the practice and thinking and precipitation of a lot, today will be the most core data intelligence system M, out to share with you.
The data intelligence system is called MANA, and the Chinese is called Snow Lake. The name "Snow Lake" comes from the second part of "The Three-Body Problem", "Dark Forest", the protagonist Luo Ji wanders between the starry sky, snowy mountains, forests, meadows and lakesides, until one day he found a way to solve the "three-body crisis" and save the earth in the lake. The name represents the idea of ai leading to the dream of autonomous driving.
So what exactly is MANA?
In terms of perception ability, for the current on-board camera and lidar of the perception equipment of the millile core, the core problem is how to make 1+1 achieve the effect of greater than or equal to 4. Compared with the previous standard result fusion method, the more efficient process fusion method is adopted, and the temporal and spatial fusion is added to the features of the timing, which quickly increases the perception ability.
In terms of cognitive ability, Gu Weihao believes that it is necessary to have three major elements of safety, comfort and efficiency. In terms of safety, The core of the self-developed safety cognitive model CSS is that the automatic driving system is not limited to ensuring that it does not actively make mistakes from a purely mechanical point of view, but also fully considers the understanding of the behavior of other traffic participants learned from the data and the historical experience of time-lapse; on top of the safety bottom line, learn comfortable and more efficient quantitative standards from the data, so that the automatic driving algorithm can better handle complex driving scenarios and formulate driving strategies that are more in line with user preferences. And through the automation of scene mining, reinforcement learning, simulation engine to build a cognitive intelligence closed-loop system, continue to extract knowledge from massive human driving data, and quickly iterate the ability of vehicle-side cognitive algorithms.
Milli is studying an end-to-end simulation learning, which is to use past examples as a guide to get specific local car actions from digital scenes. In this process, all the movements have been marked by people themselves as they drive themselves. Millimu selects the driving behavior of drivers who are more in line with the requirements and continues to train in different scenarios. At the same time, many deep reinforcement learning methods have been practiced, and a closed-loop automatic labeling system has been constructed, and an unsupervised automatic labeling algorithm has been used, which has greatly improved the efficiency of data labeling to meet the needs of large-scale mass production.
At the level of simulation capabilities, The simulation system is compared to the "autonomous driving metacosm", and the efficiency is greatly improved by verifying the effect of perception and cognition in this "metacosm".
Finally, at the level of computing power, Gu Weihao judged that in the future, driven by intelligent cars, human recorded data is shifting from text to image, and the storage and computing scale of images will dominate, which will bring new revolutions to storage and computing. Gu Weihao announced at the scene that the MANA supercomputing center is in preparation, mainly for the data processing, training, reasoning and verification needs of automatic driving, and China's automatic driving has entered the era of supercomputing centers.
One more thing: NoH in the city of Noh will be available in 2022
MANA is the core driving force behind the evolution of all millimeter abilities. Through the super capabilities created by MANA, the self-driving products of the millima are becoming more powerful and leading. Gu Weihao said that in the middle of 2022, the HPilot assisted driving system will soon launch a new function of "City NOH", which uses the world's largest computing power assisted driving domain controller small magic box 3.0, and specially optimized for the city's diverse and complex road conditions, with faster and more timely perception and response capabilities. HAOMO AI DAY also showed the current noH vehicle road test video in the end of the city, which showed many convenient functions facing the complex traffic environment in the urban area, such as traffic light recognition, avoiding traffic jamming vehicles at intersections, avoiding turning vehicles, automatically passing through the waiting area, navigating lanes, avoiding pedestrians, driving in and out of the roundabout, turning left at intersections without protection, and so on.
Gu Weihao announced the intelligent driving roadmap for Xiaomi passenger cars: in the second half of 2022, Miller will plan to deliver all-scenario NOH, and launch a fleet with HSD (HAOMO Self-Driving) in 2023. The technical power of MANA is emerging.
Summary: Finally, let us all look forward to the future will be equipped with many noH passenger cars equipped with Miller Zhixing full-scene NOH passenger cars, fleets with Millima Zhixing HSD, and more logistics unmanned vehicles with "Milliper Manufacturing" on the streets, so that the future social travel mode will be better.