I was walking through Hall B1 of the Munich Exhibition Centre in Germany, where the 2025 IAA Mobility took place. A lot has changed in the past years for one of the world’s oldest motor shows. First of all, it is not called Motor Show anymore, as in 2021 it rebranded from Internationale Automobil-Ausstellung to simply IAA Mobility and moved from Frankfurt to a less shiny location in Munich, reflecting the change the automotive industry is undergoing.
I’m passing a massive Volkswagen booth, by far the biggest in the hall and probably on the whole show. I catch a glimpse through the gate leading inside the booth: a bar, tables full of food, and crowds of dealers, KOLs and journalists feasting on the generosity of Germany’s largest carmaker. It is clear who has a home ground.
But my destination lies elsewhere. Right next to Volkswagen’s fort, a smaller booth has attracted a significant crowd. People gather around a strikingly green, futuristic-looking sedan. It is updated P7, which had its global debut here just a few hours ago as The New P7.
The proximity of the Xpeng-Volkswagen booth is not a coincidence. In 2023, Volkswagen sought Chinese partners to accelerate its EV development cycle and found them in Xpeng, to which it invested over 700 million USD for about 5% share. Xpeng will provide its E/E platform, and both automakers will jointly develop two Volkswagen-branded mid-size pure electric vehicles, which will launch in 2026.
“Meet us in front of the SEPA platform,” posted the Xpeng representative in the group chat as we were finalising details of the interview with Tingting Yuan, also known as Candice Yuan, director of the Autonomous Driving product at Xpeng.
For Xpeng, its autonomous driving is a key differentiator in the fierce China EV competition. Nio has it’s premiumness, BYD price-performance ratio, Li Auto is the leader in range extenders, Zeekr and Lynk&Co try to attract young and hippish customers, and Xpeng has it’s ADAS (advanced driver-assistance system) and autonomous driving software. Last year, the company also started heavily promoting its AI capabilities. “Xpeng is built with AI in its DNA,” said He Xiaopeng, company CEO, during a media briefing in Munich.
In an interview with Candice, we lifted the lid on many of those topics.
Before you joined Xpeng, you worked at Alibaba, where you were responsible for business operations of unmanned logistic vehicles, Little Donkey and Big Donkey (editor’s note: low-speed self-driving vehicles and unmanned heavy-duty trucks). What experience did you bring to Xpeng from that?
It’s a big difference from Alibaba Damo Academy. In Alibaba, we did logistics, which meant no driver or passenger in the vehicle. So we didn’t have to worry about comfort. We could only focus on the safety part.
But for the human drive L2+, it is very complex. Firstly, drivers want efficiency. For example, if ADAS do the right thing, but the vehicle is slower than the driver expects, he will take over. Another example – the system might calculate that the situation is safe enough to go. But the driver thinks it is too narrow or too close, and he will take over. The third example might be that if the vehicle accelerates too fast or too slow, the driver will not feel comfortable and, again, take over.
As a result, a greater level 2 autonomous driving for passenger vehicles is more complex in some areas than the L4 we developed in Alibaba for unnamed vehicles. The driver cares about emotions and feelings, so we need to think as he thinks. We need to focus more on experience.
So there is a huge difference between developing L4 for unnamed commercial vehicles, which is actually easier in some ways than L2+ for passenger vehicles with a driver?
I can give one more example. Maybe you have already taken a ride in Waymo or another robotaxi. If a robotaxi goes the wrong way, it will continue in that direction, then turn around and follow the new navigation. But in level 2, if you go the wrong way, the driver will take over. So .. it is forbidden to go the wrong way.
What would you like to achieve in Xpeng and its autonomous driving system?
I think it is the same for every team in the autonomous driving industry – to achieve level 4 and level 5.
I thought you would say this.
Yes, everyone. But I believe Xpeng has a great advantage in doing that. I’ve been in the autonomous driving industry for more than 10 years. At the previous time, I would think, who will take more advantage of it, the automaker or the ADAS maker? Just like the Android system and the smartphone makers, or Microsoft Windows, and, for example, Dell laptops. Hardware vs software.
But nowadays it is clear that hardware and software can’t be apart. They have to be connected together. And the closer the connection gets, the faster it goes.
So Xpeng is the automaker. We have a team of more than 1000 employees just focused on autonomous driving. And we can do very well in HW and SW. We also create our own AI chips ..
…I have questions for that later. Let’s talk about lidar a bit now. Three months ago, you posted an analysis of its pros and cons. Xpeng was using lidar for seven years, and last year, you removed it from the P7 sedan and all your future models. Was it a difficult decision, and how long did it take?
Actually, from the days that we used lidar, we were continuously thinking about removing it for many years. We knew we should remove it one day. But in the past years, we had no ability to do that.
Do you mean software-wise?
Yes.
Got it.
Nowadays, we are more confident that removing lidar was a good choice. Why? Our new AI system is based on a large language model based on many data. The data are mostly short videos, cut from the road while the customer is driving.
It is a short video, like 10 or 30 seconds short. Those videos are input for the AI system to train on, and that is how XNGP is upgraded. It’s learning like this, it’s learning from every car on the road.
The lidar data can’t contribute to the AI system.
Why?
Because there is only visual input, we call it VLA – vision, language, action. Lidar data are different and can’t be absorbed by the AI system. That is why our system grows very, very fast, because we can train it on so much road data.
So, when there is an accident or some problem on the road, the AI model will learn from it and update itself?
Not only from bad cases, but also from good cases. We call it good driver behaviour: You can turn lef,t you can turn right, there are always thousands paths you can take. But which is the best way? We will see how good drivers handle it and train the system from them.
I would like to learn more about XNGP (Tesla FSD alternative, editor’s comment). I did research about it, but I’m still confused about some parts. Every Xpeng car in China is now equipped with XNGP?
Yes
Is it independent entirely from HD maps, or is there some hybrid solution?
There are no HD maps. Xpeng is the first automaker to remove HD maps from it’s NOA system
Last year, He Xiaopeng said that in 2025, you will begin the rollout of XNGP globally. How does it go?
We are the first to have approval for remote autonomous parking in Hong Kong. This is the first step for global rollout.
In Europe, we already have ACC and LCC (adaptive cruise control and lane centring control, a system that helps keep the vehicle in the centre of its lane, editor’s note) for all vehicles, so we remove some burden from customers, and they can relax. It is a basic ADAS. We don’t have XNGP in Europe, and it is connected to policy.
We follow the principles of EU policies like the DCAS (Driver Control Assistance Systems) regulation.
If the policy is ready for us to bring our system here, our large model can be quickly applied anywhere in the world.
So the policy is the biggest hurdle to bringing XNGP to Europe?
I think it’s the most important reason.
Now the question everybody is probably asking you – when will you bring XNGP to Europe?
I can’t say, it depends on policy progress. But we are very eager to bring XNGP to the EU.
How many people are working at Xpeng’s autonomous driving team?
More than 1000. We have hubs in China, the US and Europe. The major team is in China, and overseas is several hundred.
How many years do you think until L5 is achieved?
L4 is very similar to L5. L4 means that in most areas you can drive it automatically. If we achieve L4, L5 is very close. The most important step is L2 to L4.
To reach L4 depends on many conditions. First, it is policy. The government policy influences it very much. In China, L4 and L5 are only approved in some small restricted areas.
China is the first market which opened to L2+. L2+ means the XNGP can drive anywhere when the driver is ready to take over every minute. So for L2+ policy, I think China is the most ahead. So if we can see any country with an open attitude to L4 and L5, the progress will be faster.
The second thing is software ability. Xpeng is very close to achieving L4. Just today (Sept 8, editor’s note), He Xiaopeng announced we will start mass production of L4 robotaxi vehicles next year.
Do you think about using an imaging radar like Mobileye?
For now, we rely mainly on the camera. But we do not set limitations. However, as of now, there is no discussion about adding this sensor.
We follow the first principle – people drive, using eyes, hands, and feet. They are not Superman who uses lasers. So we will follow this.
You developed your own ADAS chip called Turing AI. What was the reason for it?
When developing XNGP, we found that we had to connect the hardware and chip and the software more closely. We needed to provide specific requirements for the chip, but third-party suppliers were unable to redesign it to meet our specifications. Our in-house developed chip will be more usable for our system, which will help us go faster.
I would conclude with questions, what is your daily driver, but from your previous interviews, I know it is G9.
Yes, because I love big SUVs. Maybe I will change it next year for a much bigger one.
Wait..Xpeng has no bigger SUV than the G9.
Well, He Xiaopeng talked today about the L4 mass-produced robot car, which will launch in 2026. Maybe I will buy this one. But it’s still a secret.
I see. Thanks for the teaser. And your time
Thank you too.
Follow us for ev updates