XPeng Inc. (“XPeng” or the “Company”, NYSE: XPEV, HKEX: 9868. HK) is the first Chinese car company to start insisting on self-research in the field of autonomous driving, and completed its urban NGP fully automated assisted driving technology in 2021.
The current mainstream autonomous driving technology is pure vision and perceptual fusion. XPeng emphasises that its own autonomous driving technology is vision-based. However, the problem posed is that the strength of the XPeng’s visual perception can directly affect the safety and stability of the system.
A visual evolution of XPeng’s car
Photo Credit: geekpark.com
The original version of the G3 featured a 120-degree wide-angle CarLog camera, which is currently used as a CarLog and visual wiper awareness camera.
By the time XPeng designed the P7, it had erected a Tesla-like visual driving perception system, which includes a “trinocular camera + side front/rear camera + rear view camera”. In addition, XPeng also deployed a set of supplier solutions according to the standard L2 level, including “monocular camera + 5 millimetre wave radar + 12 ultrasonic radar + 4 surround view cameras”. In short, it is equivalent to a car with two sets of perception systems.
From the user experience, P7 is mainly used in high-speed NGP through high-precision positioning + high-precision map + vision fusion positioning.
To put it simply, firstly, Xpeng owners know the specific current location through high precision positioning, then XPeng owners get a preview of various road conditions and situations ahead of driving through the high precision map, and finally users use the vision camera to see the surrounding situation to verify with the features in the high precision map.
But this model has a problem, that is, when the high precision map data defects and positioning problems, the vehicle will appear to have a very outrageous driving trajectory error.
The vehicle in the NGP driving process is actually highly dependent on the fixed high precision map and positioning, and vision in which plays a minimal role, otherwise it is almost impossible to appear the camera can not identify the vehicle driving line.
The XPeng P5 is equipped with two LIDARs. These two LIDARs are used as redundancy and the added two LIDARs output 3D depth information of the surrounding environment to make up for the lack of visual perception and enhance driving safety.
In addition, the two LIDARs help XPeng accelerate the progress of urban NGP, which according to the news on October 24, 2021, is likely to be released in some cities in Q1 2022, and XPeng can then become the “first brand to achieve urban assisted driving”.
Problems with XPeng’s self-driving technology
Complexity of urban roads conditions
Urban road conditions are significantly different from highway, urban road conditions are more complex, and it is unknown how well experience can be achieved by relying on positioning and high precision maps due to many reasons such as lack of accuracy and high degree of urban road changes from the road.
In addition, perception on urban roads is very difficult in the Chinese scenario, because it needs to not miss any obstacle during the whole driving process, and clearly identify and predict the trajectory of each obstacle, and also has to comply with traffic rules to plan a reasonable travel trajectory.
Urban generalisation while getting a head start
XPeng will have to push the city NGP before Tesla launches the FSD Beta version in China, in order to maintain the function first-mover advantage.
Tesla’s data centre in China is already under construction and some functions are already in use, so if Tesla completes its regulatory filing in China, then pushing the FSD Beta version will be soon.
XPeng also has to grab with Tesla the ability to generalise the autonomous driving system in various scenarios in various cities, which is how XPeng will subsequently extend the capability of the urban NGP to all cities. At the moment, it is doubtful how many cities XPeng can generalise in relying on high precision maps.
The difficulty of reusing assisted driving modes between XPeng’s vehicles
At present, it seems that XPeng has adopted three types of assisted driving structures simultaneously, namely, the semi-self-research G3, the self-research and supplier combination P7, and the self-research-based P5.
The three types of assisted driving hardware are very different from each other, making interconnection and reusing technologies very difficult, especially when it comes to the backward compatibility of higher-order technologies.
Since XPeng’s cars each have their own assisted driving systems, and there is very little that can be reused between them, XPeng’s automatic assisted driving team has a challenging task to maintain and develop XP 2.5, 3.0, and 3.5 simultaneously.
In such a model, it will actually have a negative impact on user experience. There are some XPeng car users who say that the next one will always be better, this one will be eliminated after buying, so what about the company’s original promised “continuous growth”?
In the future, XPeng plans to put its self-driving technology into a closed-loop technology stack, similar to Tesla, so that its products can benefit from technological improvements, which are imminent.