Xpeng Motors mass-produces end-to-end autonomous driving large-scale model and launches "AI driver" globally

On May 20, Xpeng Motors held a 520 AI DAY press conference with the theme of "Opening the AI ​​Intelligent Driving Era", sharing the latest progress of AI technology and announcing the full push of the AI ​​Dimensity system to users. The push of the AI ​​Dimensity system will cover all models of Xpeng Motors on sale, and will be completed from 20:00 on May 20.

At the press conference, He Xiaopeng, Chairman and CEO of Xpeng Motors, also announced a major news: " The first model of the MONA series will be unveiled in June this year", and clearly pointed out Xpeng Motors' new market positioning - the global pioneer and popularizer of AI smart driving.

End-to-end large-scale models are put into mass production, and Xiaopeng Motors opens the era of AI intelligent driving

At present, the Chinese automobile market is undergoing a transformation from the new energy era to the intelligent era, and AI technology is the leader of this wave of technological change. AI technology will subvert the original evolutionary iteration speed and iteration method of smart cars. By learning human wisdom and imitating human thinking and decision-making logic, it will replace the traditional manual handwritten algorithm rule iteration, and enter the AI ​​era of accelerated evolution of smart cars in an end-to-end data-driven algorithm iteration mode.

The AI ​​smart driving era has given birth to AI smart driving cars. For Xiaopeng Motors, AI smart driving cars have three core characteristics, namely active learning, rapid growth, and thousands of faces for thousands of people. The core is to improve the understanding, perception and data decision-making capabilities of complex scenarios through data feeding and training of large AI models with unlimited rules. At present, Xiaopeng Motors already has the core algorithm research and development capabilities of perception, positioning, planning, and decision-making, as well as the data processing and analysis capabilities of the vehicle and cloud. It can realize the rapid iteration of algorithms based on actual data, and continuously provide users with higher levels of autonomous driving capabilities through OTA.

In terms of intelligent driving, Xiaopeng Motors released the first end-to-end large model for mass-produced vehicles in China: neural network XNet + regulatory control model XPlanner + large language model XBrain.

The three-in-one deep visual perception neural network XNet, by aggregating dynamic XNet, static XNet and the industry's first mass-produced pure visual 2K occupancy network, can make the autonomous driving system like naked-eye 3D. The industry's first mass-produced 2K pure visual occupancy network reconstructs the world with more than 2 million grids, restores the traversable space in the real world with high 3D fidelity, and clearly identifies every detail of static obstacles, which increases the perception range by 2 times to an area of ​​1.8 football fields. It can accurately identify more than 50 targets, allowing users to see more clearly and farther while driving, just like having eagle-eye vision.

In addition to perception, XPlanner has also launched a large neural network-based planning model - XPlanner. XPlanner is like the human cerebellum. Through massive data training, driving strategies are constantly evolving towards anthropomorphism, with "experienced driver-like footwork", which reduces forward and backward setbacks by 50%, illegal parking and jams by 40%, and safety takeovers by 60%, bringing users' comfort and safety experience to a new level.

After the introduction of the AI ​​large language model XBrain architecture, the autonomous driving system has the understanding and learning ability of the human brain, and the generalization processing ability to handle complex and even unknown scenarios has been greatly improved, as well as the reasoning ability of the macro logic of the real physical world. With the support of XBrain, the autonomous driving system can recognize waiting areas, tidal lanes, special lanes, road sign text, and understand various prohibitions, fast and slow behavioral instructions in seconds, and then make anthropomorphic driving decisions that take into account both safety and performance.

Based on video training with over 1 billion mileage, real-car testing with over 6.46 million cumulative kilometers, and simulation testing with over 216 million cumulative kilometers, Xpeng Motors' end-to-end large model can achieve "two-day iteration", and its intelligent driving capability will increase 30 times in the next 18 months. In the third quarter of 2024, Xpeng Motors' intelligent driving will be able to "drive across the country and on every road", and in 2025, urban intelligent driving will be comparable to highway intelligent driving experience.

In addition, He Xiaopeng also mentioned at the press conference: "In 2025, Xiaopeng Motors will achieve L4-level intelligent driving experience in China", and Xiaopeng Motors is testing the end-to-end capabilities of XNGP globally, and intelligent driving technology is beginning to go global.

The end-to-end big model released earlier once again verified Xpeng Motors' systematic capabilities in AI smart cars. In order to continue to evolve, Xpeng Motors also made it clear that in 2024, it will invest 3.5 billion yuan in intelligent research and development in "AI technology with intelligent driving as the core", recruit 4,000 new professionals, and invest more than 700 million yuan each year in computing power training in the future.

AI Dimensity System, the First Step to Open the Era of AI Intelligent Driving

The AI ​​Dimensity system, which was fully pushed at this conference, is another pioneer of Xiaopeng Motors in the field of AI technology application. Yu Tong, head of Xiaopeng Motors' intelligent experience, said that the AI ​​Dimensity system is the industry's first operating system that fully applies AI technology to smart cockpits and smart driving, and its advanced nature ensures that users can use it for a long time without lag. After the upgrade, the average frame rate, first frame tracking, and touch response speed of the AI ​​Dimensity system have all increased by more than 30%, and the application startup speed has increased by 50%. In the user public beta, the satisfaction rate of the AI ​​Dimensity system is as high as more than 97%.

After fully integrating into the AI ​​big model, the AI ​​Dimensity system will provide users with smarter exclusive in-vehicle AI assistant services through the three core functions of AI Xiao P, AI driver, and AI bodyguard, enabling users to have a safer and more comfortable driving experience regardless of whether it is intelligent driving or human driving.

In terms of smart cockpits, Xiaopeng Motors has launched the industry's first "global large language model", which deeply integrates the reasoning capabilities of the AI ​​large language model into all vehicle use scenarios. By linking the in-vehicle perception system and the external environment, it provides users with an intimate in-vehicle assistant AI Xiao P that can actively perceive and reason about their needs.

Based on the deep application of AI big language model, the new AI Xiao P will have a higher level of understanding and demand reasoning ability. For example, Xiao P can complete the cockpit settings with just one sentence, such as "When I pass this place, please remind me to pay attention to the speed" or "I will take a short nap and wake me up in 20 minutes", and AI Xiao P can provide precise services. Xiao P's reasoning can understand the emotional needs through the user's feelings. With a sentence "I'm a little cold", it can understand whether the user needs to turn on the air conditioner.

The privately customized AI driver has three powerful functions: AI driving, AI parking, and AI bodyguard, which comprehensively covers all scenarios including driving and parking, allowing the vehicle to have insight into the external environment, providing real-time information to the driver, and ensuring the safety and convenience of driving.

While the public is still discussing the "feasibility" of autonomous driving, Xiaopeng Motors has already set its sights on the future of "personalized experience". The world's first AI driver is Xiaopeng's first step towards "personal customization" and personalized services for each individual.

AI driver is also the industry's first mass-produced "point-to-point" ultra-long memory pilot intelligent driving. Based on its powerful autonomous learning ability, AI driver only needs to learn once, and the system can generate a customized driving route, providing users with an intelligent assisted driving experience covering the entire journey. A single user can store up to 10 memory routes, and each route is up to 100km long. AI driver can already achieve "0 takeover and 0 downgrade of commuting routes", truly making the intelligent driving experience comparable to human driving.

In the face of parking problems such as narrow parking spaces, Xiaopeng Motors has also launched the world's first mass-produced remote-free parking and welcoming exit function - AI parking, which can easily achieve fully automatic parking and multi-scene high-speed parking, and can increase the number of parking spaces by 3 times, and the parking speed by 50%. It can also park in complex and tricky parking spaces such as side parking and end parking. With the help of the mobile phone APP, you can pause and continue to complete the parking at any time, and the comfortable parking allows users to park "faster, better and cheaper".

"No AI, no intelligent driving." Becoming the "pioneer and popularizer of AI intelligent driving cars" is not only a new declaration of Xiaopeng Motors, but also the answer sheet that Xiaopeng Motors will give for the next decade.