Author: Chris Zheng

On June 8th, Xpeng Motors held an experience event for Xmart OS 2.6.0 in Beijing. A day before this event, Xpeng promoted VPA (Valet Parking Assist) in its poster for version 2.6.0 as the “strongest automatic parking function on the market“, which sparked our curiosity.

Valet Parking Assist refers to a system that assists drivers in parking their vehicles at a predetermined destination along a pre-set route (the route must be memorized by the system upon initial parking). During our experience with VPA, we witnessed its intelligent driving decisions such as minor adjustments to avoid obstacles, U-turns, interaction with other vehicles, automatic warning lights at intersections, and pedestrian avoidance. Afterwards, the system automatically parked the car into the available parking spot.

According to Xpeng, VPA achieves centimeter-level real-time positioning accuracy at indoor parking lots through a semantic map and matching algorithm without reliance on restructured parking lots or navigation maps from map providers, even in the absence of GPS signals.

Currently, a single Xpeng vehicle equipped with VPA can learn up to 100 parking routes, with a maximum learning distance of 1000 meters per route. This feature satisfies the parking demands of various scenarios such as residential areas, offices, and shopping malls.

After the experience event, Xpeng held a media communication meeting for Xmart OS 2.6.0. The following are the questions and answers from the session, summarized by Class Star Channel:

Q: Can VPA recognize outdoor parking lots or parking structures?

A: VPA is ultimately to be equipped to cover all parking scenarios, starting with underground lots, moving onto ground-level lots, and later, more complex structures such as parking garages.

Q: Will VPA have the functions of multi-level parking and self-summoning?

A: A demo version of cross-floor parking already exists, and the summoning function is also in the planning stages. The ultimate goal is to achieve autonomous parking.

Q: How many cars have received the public beta for Valet Parking Assist?

A: Currently, the public beta version has been released to 400-500 cars, based on recruitment from the Xpeng user community. After further review, a full-scale release will be coming soon.

Q: How much storage space is needed for memorized routes up to 1000 meters?

A: The data volume of visual mapping for each memorized route cannot be disclosed, but it can be said that** 1000 meters and 100 parking lots do not pose a significant burden on storage space**.

Q: Can data modeled by Xpeng for smart parking be shared among multiple vehicles? If so, does that mean Valet Parking Assist can also be used at non-fixed positions?

(Answer unavailable)A: Non-shared memory routes and parking spaces will not be realized by XPeng until the user agreement is reached. For non-fixed parking spaces, you can currently use memory parking to park near the parking space, and then use automatic parking to connect for parking. Of course, non-fixed parking spaces have been planned for.

A: Non-fixed parking spaces will be available soon. If there are no pedestrians or oncoming vehicles, the lights will flash to prompt drivers to proceed. If there are pedestrians or vehicles, high beam lights will be used more carefully.

A: In underground parking lots, the reflection of millimeter wave radar waves can be very chaotic, such as ceilings, pillars, and metal objects, so the use of vision is relatively high, relying mainly on front, rear and panoramic cameras for fusion.

A: When encountering another vehicle traversing at a crossroads or T-shaped intersection, the memory parking feature will exit. It only reminds the driver to take over, but does not apply the brakes.

A: Edge recognition is mainly done through panoramic cameras and ultrasonic radar, but, for now, the identification of obstacles is not yet done in the SR, so this will be updated in future versions for side front and side rear recognition.

A: Parking identification will fail if there is a wall on one side of a parking space with only one line on the other side. The training scenario has not yet covered unmarked parking spaces.

A: On a normal two-way road, if there is enough space for P7 to pass through, it should be fine. XPeng will also consider whether to limit pedestrians in case of adverse driving situations.

A: Ground arrow identification has been added to overcome the problem of driving in the opposite direction. This update will be included in subsequent versions.

A: The SR modeling has changed slightly in version 2.6.0. Which department is responsible for visual modeling?A: The SR visual signal comes from the autonomous driving team, providing information about vehicle, object type, location, attribute, etc., and then the application R&D team of the Internet center restores it in a 3D environment.

Q: Do you have to go back to the starting point of the memorized route the first time it was ran in order to enter the memory parking mode?

A: There’s no need to return to the starting point of the memorized map 100% to activate memory parking. It can be activated from any point on the route.

Q: In the process of learning the memory parking, the first step is to learn the route for the first time, and the second step is to leave the parking space about 200 meters before it can be activated again. What considerations led XPeng to choose to leave around 200 meters?

A: The reason why it is necessary to leave 200 meters before restarting is that it is helpful for the recovery of GPS signaling when returning to the ground and repositioning on the road. After entering the parking lot, the GPS signal is lost, and all positioning is based on semantic maps.

Q: Currently, can memory parking not allow passengers to get out of the car because it is technically unachievable, or because it is safe not to do so?

A: Safety and regulations are serious restrictions on memorized parking. Memory parking is not autonomous driving and does not aim to replace drivers. There is still a long way to go for unmanned driving. The problem with summoning is the same.

Q: Can memory parking be used to find XPeng’s official charging station?

A: XPeng has also planned for supercharging memory parking. It’s a bit like NGP of the parking lot. The user does not have to find it the first time he goes to XPeng’s supercharging station, and it can be completed through memory parking.

This article is a translation by ChatGPT of a Chinese report from 42HOW. If you have any questions about it, please email bd@42how.com.