Author: Cherio Cheng Yuan

On October 24th, 1024 is the Programmer’s Day and also the technical day of XPeng Motors. (Hereby I wish our CTO Meow a happy holiday)

At the beginning of the press conference, He XPeng hoped to change human life through technological innovation, which was his original intention of working and starting a business. At the same time, he also revealed some information that starting from XPeng G9, including later models, super fast charging will be fully popularized.

In addition, He XPeng also has some personal feelings about the urban NGP, high-level intelligent assisted driving. At the beginning of using urban NGP, he complained that this was a novice driver. But lately, he felt completely different when using the fourth “internal version”, from “belief” to “physical sensation”.

Of course, the urban NGP also faces many difficulties, such as the management of high-precision maps. How to solve this problem next year, XPeng will answer it at this 1024 by Dr. Wu Xinzhou.

Intelligent Driving: When NGP Has “Physical Sensation”

As the Vice President of XPeng Autonomous Driving, Dr. Wu Xinzhou believes that by 2023-2024, high-level intelligent driving will cover more and more cities.

Many people will ask, why does XPeng want to do the city scene that is more than a hundred times harder? Facing the complex roads, irregular traffic behavior, and frequent lane changes and construction, this is a super complex and constantly changing scene.

Wu Xinzhou said that XPeng insisted on “first principles” in building urban scenes, as it is the most frequently used scenario for users. In the process of using a car, city roads account for 71% of the mileage and 90% of the time, and 100% of the users pass through city roads when driving.

This is the key to XPeng’s breakthrough in intelligentization. They need to make users willing to pay for intelligence.

As we all know, the complex NGP scene in the city requires stronger software capabilities. Dr. Wu Xinzhou gave some rough data, such as a 4-fold increase in perception model quantity, an approximate 6-fold increase in code volume, and a planning and control code volume that increases even as much as 88-fold.

Dr. Wu Xinzhou said that XPeng’s urban NGP can cope with complex traffic environments with ease, such as lateral control, game theory, and uncertainty scene processing. It can use maps with “mastery” – when encountering static roadblocks, temporary detours, and large intersections, much of the information in high-precision maps is not applicable. At such times, the urban NGP will temporarily abandon the high-precision map information and automatically downgrade to ACC+LCC.

In fact, for urban NGP, high-precision maps like “light rail” do not work. XPeng has provided some internal data on this.In the NGP testing users who have already received the push, the penetration rate of city ADAS reaches 84%, and the penetration rate of city NGP mileage reaches 63%. Compared with human drivers, the travel efficiency is 90%, and the take-over rate per hundred kilometers is 0.6.

For XPeng Intelligent Driving’s roadmap, full-scenario ADAS XNGP is the “ultimate form” before autonomous driving. XPeng will go all out towards autonomous driving after 2025.

Dr. Wu Xinzhou said that the goal of XNGP is to “catch up with, or even surpass, veteran drivers”. What can compete with veteran drivers? The answer is XNet, a new generation of perception architecture that features the keywords of multi-camera/multi-frame.

In terms of static perception, it has strong environmental perception ability and generates “high-precision maps” in real-time.

In terms of dynamic perception, it has stronger 360-degree perception, better game skills, and higher successful lane-changing rate. The purely visual speed perception and intent prediction are a bit like “Observation Haki” in One Piece.

In fact, behind XNet is a massive amount of data labeling, which of course needs to be rolled up. XPeng uses fully automatic labeling system, which improves efficiency by nearly 45000 times. In addition, XPeng’s “Fuyao” is China’s largest self-driving intelligent computing center with 600 PFLOPS of computing power.

Finally, Dr. Wu Xinzhou elaborated on many specific technical implementation details of XPeng’s software engineering. If you’re interested, you can watch the conference replay, which is not repeated here. In short, all of XPeng’s software work is for better intelligent driving services.

When will XPeng’s new intelligent driving capabilities be available for Pengyou? Dr. Wu Xinzhou gave two timelines.

The XNGP release plan is shown below:

The XPILOT release plan is shown below:

One more thing: XPENG Robotaxi is coming! XPENG will gradually open its Robotaxi operations to the public, proactively identifying operational difficulties, and implementing a closed-loop data feedback system to enhance its level of assisted driving capabilities.

Dr. Wu Xinzhou has been with XPENG for four years now, believing that he arrived at the right platform at the right time, and that the team is on the right path to “free humanity from driving.”

Intelligent Interaction: Possibly the Ultimate In-car Dialogue

Regarding intelligent interaction, XPENG mainly demonstrated the capabilities of its full-scenario voice 2.0 system.

Full-scenario voice 2.0 is the first in-car voice system in the industry to apply MIMO multi-audio zone technology. It can accurately process four channels of audio at the same time, support contextual dialogue across different audio zones with an accuracy rate of over 96% for multi-user ambiguous semantic identification, and support multi-channel audio feedback.

XPENG CEO He XPeng stated that Full-scenario voice 2.0 is highly responsive, exemplified in the following scenarios:

1. Listen, think, act, and answer simultaneously;

2. Wake-up to interface feedback takes only 245ms, while voice feedback is less than 700ms;

3. Commands are executed within a delay of 1 second.4 physical commands executed instantly with a single sentence.

Moreover, the intelligent robot provides all-scenario offline voice service, supporting over 600 functions with stable end-cloud integration.

Intelligent Robot: Heading to Mass Production

The quadruped robot “Little White Dragon” has evolved with the addition of a six-axis robotic arm “tail” version and has gradually transitioned from its “experimental state” towards mass production. This robotic foal now possesses the following functions:

1. AR projection: exploring innovative interaction;
2. Flexible textile joint: protecting motion safety;
3. Liquid silicone skin: impact and scratch resistance;
4. EVA supercritical foam insoles: shock absorption, noise reduction, and wear resistance;
5. High-fidelity sound system: delicate verbal communication and multimedia entertainment;
6. Six-axis robotic arm: expanding multi-scenario services;
7. Lightweight skeleton: light and sturdy.

Throughout the process of developing the quadruped robot, XPeng (He XPeng) has considered and explored various issues, including the tranquility of the robot’s movements. XPeng mentioned that an unoptimized quadruped robot can produce a 70-decibel noise while walking at home. If we genuinely want to have four groups of robots enter our households, the noise issue must be addressed. Therefore, XPeng utilized EVA supercritical foam to provide shock absorption and noise reduction to the robot’s soles.

Narrow Environment Obstacle Avoidance

In a household indoor environment, obstacles are always difficult to avoid, and the traditional way of simply moving forward and backward to avoid obstacles is clearly impractical in a limited space. In other words, robots need to avoid obstacles more like living beings.

Multi-Scene Application of Robotic Arms

Because quadrupeds need to land, the interaction of this type of robot relies more on additional robotic arms. Currently, it has been able to perform daily actions such as opening doors, turning on lights, and erasing blackboards.

XPeng applied the automobile R&D system to the field of robots and squeezed vehicle engineering technology into smaller robot bodies, which is indeed a technical challenge.

Flying cars: XPeng takes off seriously

Previously, XPeng Huitian Flying Car made its overseas debut. XPeng believes that a flying car should have features such as road-air integration, vertical takeoff and landing, electric, and intelligent.

Next, the most shocking hardware product in this press conference appeared. XPeng unexpectedly showed a real “flying car,” a split car with a body structure separating the upper and lower parts. I’ll do my best to describe it to you: the lower half is still a car, and there is a foldable four-rotor arm large drone propeller on the top of the car roof.

“`markdown
To be honest, I thought it was just a rendering video at first, but unexpectedly, XPeng is really doing it. The internal code name is X3, which made a two-ton car fly and the cruising range can reach tens of kilometers.

Alright, no kidding, XPeng Huitian Flying Car is basically formed and more mature. This flying car adopts a distributed multi-rotor configuration and has the following advantages:

1. Safety advantage: Equipped with fault-tolerant control function of flight control and power dual backup;

2. Use advantage: Smaller overall size, smaller take-off and landing site restrictions, better flight passability;

3. Structural advantage: Reduced production difficulty, simple folding, safe and reliable;

4. Noise advantage: Lower noise level;

5. Experience advantage: Adopt the same configuration as the exploration product and accumulate rich design and experimental experience.

In addition, the self-developed flying car driving and control system of XPeng Huitian has two modes: airborne and ground. When in ground mode, it is consistent with the driving mode of a traditional car. When in airborne mode, it can easily achieve movements such as forward, backward, turning, ascending, hovering, and descending by coordinating the steering wheel and gear lever.
“`

Driving to the sky, XPeng is serious!

Well, let’s summarize XPeng’s future and present.

In terms of intelligent automobile technology, XPeng once again maximizes its advantages.

With the cooperation of Pengxing robots and Huitian flying cars, XPeng has also bid farewell to the attribute of a single automobile company, showing the ability to provide a complete set of future-oriented travel solutions.

The only remaining question now is how to make automobiles a hot-selling product.

-END-

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.