Editor? /Huang Dalu?
Design? /Zhao Haoran
With the advanced intelligent assisted driving entering the second half, the wind direction in the field of automatic driving has changed suddenly recently.
For a long time in the past, Tucki, Ideality, Weilai, Huawei and other enterprises used high-precision maps as the bottom information of driving decisions. However, from the second half of 2022, the autopilot circle began to move towards high-precision surveying and mapping. After years of layout, many enterprises gave up the route of pure high-precision surveying and mapping and changed lanes one after another.
Especially in the process of intelligent driving infiltrating into cities, this wave of "de-high-precision maps" reached its peak in early 2023 due to the influence of policies.
On this track, a company has been running for many years.
In China's self-driving rivers and lakes, Zhixing was the first to put forward the route of "re-perception", and it was almost the only manufacturer in the industry to take this route for quite some time. In the highly competitive market, predicting the right direction in advance means taking fewer detours and winning valuable time. The same is true in reality.
On April 1 1, the 8th Mo Hao AI Day, Mimo officially announced the designated cooperation of the three major OEMs again. The brand-new 2023 tank 500 equipped with Mimo HPilot2.0 was launched in March. The first brand-new DHT- equipped with HPilot3.0 will be launched soon, and the second model equipped with Mimo HPilot3.0 will also be in Weibai Blue Mountain.
At this point, Millie has completed an important step of commercialization leap, showing Millie's unique speed.
As the only AI technology feast focusing on autonomous driving in China, every AI Day in Mo Hao has become an industry event. This year's AI Day once again brings together the ultra-luxury guest lineup in the field of autonomous driving, which not only has the most cutting-edge technology release, but also has profound insights.
"2023 will be the year of smart driving sprint and the year of big exams." At the event site, Zhang Kai delivered a speech entitled "Mo Hao Speed, ai Speed".
He pointed out that in 2022, the front-loading rate of L2 and above assisted driving passenger cars in China market will rise to 29.40%, and 5,859,900 vehicles will be delivered up to standard. Last year, it was predicted that by 2025, the loading rate of advanced assisted driving will exceed 70%, which means that smart driving products will enter a full-scale outbreak period of rapid growth, and 2023 will be a very crucial year.
First of all, in 2023, urban navigation-assisted driving products will be mainly mass-produced, and the urban assisted driving products of major players will enter the competition of real user coverage and multi-city landing. Secondly, the integration of parking and navigation and the commercialization of unmanned vehicles will become the focus of autonomous driving companies. In the field of passenger cars, smart driving products equipped with integrated parking and parking functions will usher in a wave of pre-installed mass production; In the unmanned field, terminal logistics automatic delivery vehicles have exploded in supermarkets, express delivery and other scenes, and in 2023, a closed loop of sustainable commercialization will be realized in these scenes. Large-scale models and intelligent computing centers have become the new infrastructure for autonomous driving.
If we compare these four fields to four battlefields, we have launched four major battles in them.
The controversy about accelerating landing has a long history, and there is no theory at all. It is necessary to practice the automatic driving route.
When many car companies re-emphasized the map route, in April 2022, Mimo Zhixing said that it would do urban intelligent driving with heavy perception, reduce the weight of high-precision maps in intelligent driving, and even do not need high-precision maps. The reason for this decision is that it is judged that the update frequency of high-precision maps is too low to meet the needs of intelligent driving.
Gu Haowei, CEO of Mimo Zhixing, once made an image metaphor for this: high-precision maps provide road transcendental information, but they are unstable and have been under construction. If you regard it as a lidar, and this lidar can never be constructed, its confidence will certainly decrease.
The following story has proved that the re-perception route is a commercial road for autonomous driving. Foresight has enabled China to speed up the mass production of autonomous driving, and its city NOH has become a mass-produced auxiliary driving system because of its strong perception.
At the beginning of 2023, the battle for the king of intelligent driving capacity started. Today, there are nearly 20 third-generation passenger car products, and the new Moka DHT-PHEV will be the first product equipped with HPilot3.0; The mileage of millimeter intelligent assisted driving users has exceeded 40 million kilometers, and the city where millimeter NOH is located has begun to promote testing in Beijing, Baoding, Shanghai and other cities, and you can experience the function of millimeter NOH when you land. In 2024, the total number of nano-assisted passenger cars is expected to reach the order of one million.
Overseas markets are also accelerating their layout. Vehicles equipped with millimeter HPilot will be delivered to users in the European Union, Israel and other regions and countries, and the Mexican and Russian versions will be mass-produced; Next, it will be launched in the Middle East, South Africa and Australia markets to help China drive intelligently to the sea.
When it was overwhelming, the "NOH Hundred Cities Battle" was detonated at the last minute. At the event site, Zhang Kai played a video of road test, which completely recorded the scene that NOH, a city in Beijing and Shanghai, could turn smoothly at the intersection of traffic lights and control cars, and realized an unprotected left turn.
In order to speed up the attack on the city, the principles of safety first, users first and scale first are put forward, and priority is given to landing cities that are standardized, cities that users expect, and cities with large sales of mounted models. With the gradual landing of mass production models, it is estimated that NOH will land in an orderly manner 100 cities by 2024.
For advanced technology, application is far more important than theory. It is more important to get through specific scenes and mass-produce intelligent driving systems than to simply emphasize a higher level of autonomous driving. On the one hand, consumers can experience the convenience brought by autonomous driving as soon as possible, and at the same time, they can accumulate valuable data and experience for technological development.
With vision, courage and strength, I can be called a practitioner who dares to think and do. It can be expected that with the landing of more production models equipped with millimeter HPilot intelligent driving products, millimeter will continue to consolidate the leading position of mass production automatic driving and have more imagination in the future.
The appearance of Hai Ruo reshaped the technical route of automobile intelligence, and artificial intelligence technology is welcoming a major node from quantitative change to qualitative change.
In 2023, the popularity of ChatGPT applications and the release of large-scale models such as GPT-4 made everyone believe that an era that truly belongs to artificial intelligence has begun. Some people say that the emergence of ChatGPT is no less than the invention of PC and Internet. Others say it's AI's iPhone moment.
In any case, the arrival of ChatGPT is about to change mankind, and the future has arrived. Millicent Zhixing launched the world's first self-driving generation model-Drive GPT.
In his speech, Gu revealed the origin of the Chinese name "Snow Fox" of DriveGPT: "Hai Ruo" comes from "Autumn Water in Zhuangzi", in which there are two mythical figures, Hebo and Beihai Ruo. Hebo asked Beihai, What is the difference between size and size? If Beihai teaches him, he will not feel great because of heaven and earth, nor will he feel small because of nuances.
In the end, DriveGPT was named Snow Fox Hai Ruo, which symbolizes wisdom and tolerance and contributes to the development of the industry.
As the world's first large-scale model of autonomous driving, Snow Fox Hai Ruo introduced driving data to establish RLHF (Human Feedback Reinforcement Learning) technology to continuously optimize the cognitive decision-making model of autonomous driving. Taking out products consciously in the whole industry is due to the research on large models from a very early age.
Hai Ruo's goal is to achieve end-to-end autonomous driving. At present, it is mainly used to solve the cognitive decision-making problem of autonomous driving, and will integrate the capabilities of many large models in the future.
At present, Snow Fox Hai Ruo has upgraded its model architecture and parameter scale, with the parameter scale reaching 654.38+0.20 billion. In the pre-training stage, the driving data of 40 million kilometers of production vehicles were introduced, and in the RLHF stage, 50,000 manually selected difficult scenes were introduced to take over the editing.
In addition, Snow Fox Hai Ruo has also opened up the scene recognition ability, which can train the model according to the tips of the input terminal and the decision samples in the CSS autopilot scene library, so that the model can learn the reasoning relationship, thus splitting the complete driving strategy into the dynamic recognition process of the autopilot scene, and completing the generation of an understandable and interpretable reasoning logic chain. This evolution has greatly promoted the progress of the whole industry.
At the scene, Millie announced that the first model of Snow Fox Hai Ruo is the new Mocha DHT-PHEV which will be mass-produced. This will make the city's NOH capability lead the industry for at least one year, and the emergence of DriveGPT will reshape the automotive intelligent technology route.
Gu mentioned that Snow Fox Hai Ruo can be gradually applied to urban NOH, quick recommendation, intelligent sparring, rescue and other scenes. With the blessing of Snow Fox Hai Ruo, the vehicle will be safer; The action is more humanized and silky, and there is reasonable logic to tell the driver why the vehicle chooses such a decision-making action. For ordinary users, vehicles are more and more like old drivers, and users will have a stronger trust in smart products and understand that the behavior of vehicles can be predicted and understood.
"Millimeter is the most self-driving company that knows mass production and landing. It is the most self-driving in Tier 1 and the most closed-loop data among companies with data." Zhang Kai expressed some pride at the scene.
The pace along the way, the understanding and utilization of data, the use of the largest intelligent computing center of autonomous driving company, and the launch of the brand-new Snow Fox Hai Ruo have further strengthened the advantages of Mimo in data closed loop. With the launch of more production cars, the advantages of Mimo may be further amplified.
Based on technology, Millie sprinted to the 3.0 era of autonomous driving. On the 6th Mo Hao AI Day in September, 2022, Millie first mentioned the 3.0 era of autonomous driving in the industry.
Combined with the law of technology development and the development history of autonomous driving in recent ten years, I think the development of autonomous driving technology in recent ten years can be divided into three stages: the earliest hardware driving mode is the era of autonomous driving 1.0; In recent years, the software-driven mode is the era of automatic driving 2.0; It will happen soon and will continue to develop data-driven methods for the autonomous driving 3.0 era.
Therefore, Millie has been preparing for the era of autonomous driving 3.0. In terms of perception, cognition and pattern construction, they are all built in a data-driven way. MANA, the first automatic driving data intelligent system in China, was built by AI DAY. After more than a year of application iteration, it also ushered in a comprehensive upgrade and officially opened up empowerment.
Gu Hao Wei introduced that MANA computing basic services are specially optimized for large-scale model training in terms of parameter scale, stability and efficiency, and integrated into OASIS; Secondly, MANA's perception and cognition-related large-scale model capabilities are integrated into DriveGPT; Third, increase the data synthesis service using NeRF technology to reduce the acquisition cost of corner case data; At the same time, heterogeneous deployment tools and vehicle adaptation tools are optimized for the rapid delivery of multi-chip and multi-vehicle models.
& ltP class="ql- This article is from auto business review, the author of Yiche number, and the copyright belongs to the author. Please contact the author for any form of reprint. The content only represents the author's point of view and has nothing to do with the car reform.