Has Xiaopeng's intelligent driving surpassed Huawei? A comprehensive understanding

Excerpt from Yuan Guoqing

Yu Chengdong said: We are the first.

He Xiaopeng said: We are 5 times ahead of the industry leader.

Who is bragging? Who is telling the truth?

To clarify this issue, I had dozens of conversations with AI. Today, I will explain it thoroughly from seven dimensions.

1. Route Differences

Huawei insists on a safety redundancy route of "LiDAR + multi-sensor fusion," while Xiaopeng bets on a physical AI route of "pure vision + end-to-end large model."

Behind this is actually two different sets of "brains" at work.

Huawei is equipped with a "world model," which first reconstructs everything seen by LiDAR and cameras into a 3D virtual world, and then makes decisions. The advantage is a very high baseline, ensuring no "hallucinations" of non-existent things.

Xiaopeng uses a "physical AI model," where the camera sees the scene, and the large model directly converts it into steering wheel and pedal operation commands. The advantage is extremely fast response, with a driving style more akin to humans, and the iteration speed is exponential.

In simple terms, Huawei adopts an engineer's mindset—first calculating the world clearly, then controlling it, pursuing certainty and safety boundaries; Xiaopeng adopts a biomimetic mindset—mimicking the direct pathway of human "eyes-brain-hands and feet," pursuing efficiency and human-like feel.

2. How to Perceive the World

LiDAR actively emits lasers and measures the time of echoes, obtaining physical 3D true values.

Cameras passively receive light and output 2D images, only providing width and height, with no knowledge of depth.

In the past, pure vision could not distinguish between murals and real obstacles; now the solution is "BEV + occupancy network":

BEV stitches together images from multiple cameras around the vehicle into a bird's-eye view (similar to a satellite map).

The occupancy network divides the space around the vehicle into countless tiny 3D voxels (small cubes) and uses AI algorithms to determine whether each space is occupied or has obstacles. Its core logic is "constructing 3D through reasoning rather than measurement."

This solution has extremely high requirements for models and computing power.

3. What is the perception boundary?

Mainstream 8 million pixel cameras can capture objects 200 meters away on sunny days, but in environments like heavy rain or nighttime, the effective distance is reduced.

4D millimeter-wave radar has a perception range of 200-350 meters, significantly improving the recognition ability of stationary objects compared to traditional millimeter-wave, but still has limitations for objects with extremely low reflectivity.

Common LiDAR has a perception range of 200-300 meters, with accuracy far exceeding other sensors, able to accurately identify obstacles 200 meters away, providing precious time for lane changes or braking.

4. Safety Redundancy: Will it cause confusion?

In 99% of everyday scenarios, pure vision is completely feasible.

But in 1% of extreme scenarios, such as backlighting/heavy rain/fog, the camera image blurs, making LiDAR safer;

Many people worry: Will multiple sensors conflict?

For car companies with average algorithm levels, problems may indeed arise. But Huawei's core logic is: vision dominates decision-making, and LiDAR only provides physical backup. In the vast majority of everyday scenarios, LiDAR remains "silent," only intervening when vision fails, and generally does not create confusion in daily scenes.

5. Who has faster iteration speed?

Pure vision has an unmatched advantage that LiDAR cannot compare: iteration efficiency.

Pure vision relies on software and models, and can be upgraded via OTA, with iteration cycles measured in "weeks"; LiDAR relies on hardware, requiring redesign, tape-out, and mold opening, with cycles measured in "years."

More critically, data: pure vision has a unified data format, with low collection, annotation, and training costs, forming an exponential flywheel of "more data → stronger models → higher sales → more data."

LiDAR point cloud formats are not unified, annotation costs are several times to ten times higher than images, making it difficult to form a universal data closed loop.

In simple terms: AI evolves exponentially, while hardware grows linearly. The longer the time, the more apparent the generational advantage of pure vision.

6. Who has greater market value?

I asked AI a question: If you were an investor with 100 billion funds, which route would you invest in?

AI's answer was: As an investor with 100 billion funds, the final decision is not "either/or," but "layered layout":

1. Invest 70% in the pure vision route: betting on scalability and long-term monopoly. Pure vision can achieve universal adoption of intelligent driving, occupy the largest market share, and create dual barriers of data and models, possessing platform-level value. More importantly, the pure vision model can also be applied to robots, flying cars, and other physical mobile devices.

2. Allocate 30% to the LiDAR fusion route: hedging risks and earning high-end profits. The LiDAR route follows the path of "high-end → safety reputation → brand premium," while also avoiding the risk of future policies mandating redundant hardware.

7. How should ordinary people choose?

For consumers, the choice logic is:

For urban commuting, focusing on cost-effectiveness and believing in the speed of algorithm evolution, choose Xiaopeng.

For frequent long-distance highway driving, facing complex weather, with extreme safety requirements, and not concerned about money, Huawei is the more reliable choice.

The competition between pure vision and LiDAR is fundamentally a dialogue between two technical philosophies: one side believes that AI's capabilities will eventually cover everything, while the other side believes that the physical world requires physical redundancy as a backup.

Looking back at this competition, we find that Huawei's stability and He Xiaopeng's aggression are different paths to the same peak—one paves the way with engineering, while the other climbs with algorithms.

The true victory or defeat is not about who replaces whom, but about who can bring L4 to ordinary people's lives faster, safer, and more inclusively.

Which route do you favor more? Let's discuss in the comments

#Xiaopeng Motors #Huawei #Intelligent Driving #Yu Chengdong #AI