3 million cars.
That is roughly the fleet size running on BYD’s assisted driving systems, a figure the company presented with some serious confidence this past May in Shanghai. Yang Dongsheng, a senior VP who runs the group’s tech research, didn’t just throw up some vanity metrics for the Intelligent Connected Vehicle Technology conference.
He talked about survival rates. Or at least, airbag deployment rates.
“Severe accident rates [are] reduced to one-sixth of human-driving levels,” Yang told the room.
Measured per 10 million kilometers, those airbags stay deflated when the system is active. That’s a stark reduction. Not incremental. A massive drop.
The “God’s Eye” effect
We are talking about scale here.
More than 60 models now sport this tech. Almost every passenger car in the BYD stable is offering at least Level 2 assistance. But availability doesn’t equal adoption.
Does it?
BYD says drivers are actually using it.
Navigation-assisted driving activation sits above 50%. Parking assistance hits 86%. And when you use that parking feature, you barely scrape the paint. The company claims scratches and minor bumps during parking have fallen to one-fiftieth the rate of a human behind the wheel.
Pretty hard to argue with those numbers, even if they feel suspiciously clean.
This push toward the “God’s Eye” intelligent driving system started big in early 2024 (though some reports cite 2025 for the massive expansion phase) and included a guarantee program launched back in July. It’s all part of a bigger tease. The data dropped on May 21, just a week before their main strategy event on the 28th.
Everyone was guessing. Will they roll out the new God’s Eye to more models? How deep does the integration go?
Silence on their end, loud speculation everywhere else.
Xuanji and the daily grind of data
You can’t just bolt sensors to a chassis and call it AI.
BYD leans on something called the Xuanji Architecture. It fuses electronic structure with electrification systems into one platform. A unified nervous system, essentially.
But hardware is cheap compared to training.
Here is where it gets heavy.
BYD is churning out 190,000 kilometers of simulation data every day. They use cloud-based world models. They use reinforcement learning for the weird edge cases—the “long-tail” stuff that actually causes crashes. And they update their algorithms.
Every three days.
It’s not just software logic.
They use physical AI models directly on the car. These run predictive functions, calculating defensive responses in real time. When parking, the system blends visual occupancy networks with lidar detection.
Why? To see obstacles that hang in mid-air. Or hollow shapes. Things standard cameras often miss or misinterpret as background noise.
It’s getting smarter about the void, not just the solid world.
When things go wrong fast
Simulations are tidy. Real life isn’t.
Yang’s presentation spent significant time on extreme conditions. Tyre blowouts at high speed.
Rain.
Snow.
Low-adhesion roads.
If you hit a blowout at highway speeds, the window to survive is measured in milliseconds. BYD claims their integrated platform stabilizes the vehicle within 200ms. It coordinates the motors and chassis instantly. No panic inputs. Just correction.
Their validation scenarios push tyre-blowout stability tests past 200 km/h.
They don’t just run computer codes for this stuff either. There are dedicated proving grounds. Real dirt, real steel, real rubber combining with simulation results. It’s a hybrid testing loop designed to break the car so the driver doesn’t have to.
What comes next?
The recent launches already show the pattern.
The updated Atto 3 landed in China this week with a bigger range and flash charging. The Seal 08 sedan drops later this year—second quarter—with rear-wheel steering.
More power. Faster charge.
But the real shift is the brain inside the metal.
