eastwest connect

Xpeng: XNGP technology sharing meeting in Shanghai after the Full-scenario Extreme Challenge

It is learned on April 1 from Sina Auto that on March 31, Xpeng ended the 6-day Mission X 2023 XNGP Full-scenario Extreme Challenge test drive experience event, and held an XNGP technology sharing meeting in Shanghai.

At the meeting, Wu Xinzhou, Vice President of Xpeng Autonomous Driving Center, announced that a new version of OTA — Xmart OS 4.2. 0, officially opening the first phase of XNGP capabilities to Xpeng G9 and P7iMax users.

The first phase of XNGP’s ability to open the City NGP supports the three cities of Guangzhou, Shenzhen and Shanghai

The XNGP intelligent assisted driving system is a new generation of intelligent driving system of Xpeng. It will have the ability of intelligent assisted driving in full scenarios, and it is also the ultimate product form of assisted driving before the realization of driverless driving. The complete form of the XNGP system will not rely on high-precision maps, and can seamlessly connect various scenes such as highways, cities, and underground parking lots in various cities across the country, and realize an assisted driving experience without breakpoints from the starting parking space to the ending parking space. Not only can cover daily commuting in the city, close to zero takeover on expressways and urban expressways, but also surpass most drivers in terms of safety, road game ability and traffic efficiency.


It is worth mentioning that, in addition to the G9 Max and P7i Max models equipped with the XNGP system, the Xpeng P5 P series models are also simultaneously opening NGP in Shanghai.


In areas without high-precision map coverage, the application of lidar greatly enhances the performance and application scenarios of LCC lane centering assistance, and will greatly improve the driving comfort of drivers.
The enhanced version of LCC will have the ability to go around the lane. When there are obstacles such as stationary vehicles and cones in front, and the system detects that the surrounding environment is safe and the driving state meets the conditions, it can automatically initiate a detour across the lane line.

In addition, the enhanced version of LCC will open the ability to recognize traffic lights and go straight through intersections, support the identification of the current lane direction, match relevant traffic lights, and realize red light stop; In the through lane, it supports green light start, and in the non-through lane, the system will give voice prompts. In the second half of this year, most major cities across the country will open the ability to turn left and right in no-map areas, making the user experience in no-map areas close to that of urban NGPs.

Based on the stronger perception ability of XNet, XNGP’s instrument environment simulation display has been fully upgraded. The new generation of SR reconstructs the system’s perception of the environment through 3D models, increases the types and quantities of perceivable objects, and adds motion animation effects to pedestrians and bicycles. The overall SR image is more delicate, smooth and real. On this basis, explaining the decision-making, planning and functional status of the system can not only highly restore the real environmental information during intelligent driving, but also help explain the decision-making process of intelligent driving, and improve the safety and trust of human-machine co-driving.


At the sharing meeting, Dr. Wu Xinzhou also brought a bonus: a new generation of highway NGP. It is based on the new generation technology architecture of XNGP, combined with the research and development experience brought by urban NGP, it has qualitatively improved in terms of safety, usability, comfort, traffic efficiency, and interaction methods.

Wu Xinzhou revealed that the current research and development progress of the new highway NGP is in line with expectations, and it is expected that the next major OTA version will be opened to users with the XNGP system.

Intelligent assisted driving: technical strength and system advantages

Wu Xinzhou said that the perception ability is the first step to realize navigation-assisted driving in areas without maps. XNet deep visual neural network is also China’s first and only mass-produced BEV perception. With its super environment perception ability, XNet can fuse the data collected by multiple cameras before multi-frame timing, and output 4D information of dynamic objects and 3D information of static objects, which greatly improves intelligent assisted driving, especially the ability and efficiency of perception, prediction, decision-making, and execution in the face of complex urban scenes.

Compared with the first-generation visual perception architecture, XNet uses neural networks to replace the complicated handwritten post-processing logic, and realizes end-to-end data-driven algorithm iteration. In the future, users will not be limited by the area of high-precision maps, and can enjoy the safe and efficient travel experience brought by high-level intelligent assisted driving in more cities and scenes faster.


If the super perception ability makes Xpeng take the lead in the field of intelligent driving, the strong system capability and efficiency are the core advantages and moats for Xpeng to maintain its leadership.

Xpeng has created a fully automatic labeling system, which has increased the labeling efficiency to 45,000 times that of manual labeling, and built China’s largest autonomous driving intelligent computing center in Wulanchabu, which has increased the efficiency of AI model training by 602 times.

At the same time, the ultimate engineering capabilities accumulated from the research and development of urban NGP has made it possible that the 122% Orin-X computing power required by dynamic XNet only needs 9% after optimization. The ultimate engineering deployment capability is the key to the launch of XNet, and it is also the key to continuous iteration in the future.

The ultimate engineering deployment capability is the key to the launch of XNet, and it is also the key to continuous iteration in the future.