Huawei Qiankun ADS 4

What it is

Huawei released Qiankun ADS 4, its intelligent driving system, in Shanghai on 22 April 2025.^[1] It is a combination of software architecture, sensors and version tiers that Huawei supplies to carmakers, reaching drivers through partner vehicles rather than as a product sold on its own.^[2] At the launch, Huawei's automotive chief Jin Yuzhi said 22 partner models were already on sale with Qiankun.^[2]

The WEWA architecture

The core of ADS 4 is an architecture Huawei calls WEWA — world engine plus world action model — which divides the work between the cloud and the car.^[1]

The world engine sits in the cloud and generates diffusion models of difficult cases; Huawei says the density of hard scenarios in these generated worlds is 1,000 times that of the real world.^[1] The reasoning is about scarcity. Ordinary driving footage is mostly uneventful, while the rare situations that decide whether a system is safe — a pedestrian stepping out from behind a bus, a truck stopped across a blind corner — are exactly the ones a fleet collects slowly. Generating them in bulk lets the system be trained and tested on situations it would otherwise meet only occasionally.

The second half, the world action model, runs in the vehicle. Huawei describes it as the industry's first native foundation model for intelligent driving, with all-modality perception.^[1] "Native" here contrasts with the older practice of building separate perception, prediction and planning modules and passing hand-made interfaces between them: a foundation model is trained on driving data as a whole and takes in several sensor modalities at once.

Claimed improvements

Huawei gives three figures for ADS 4 against the previous generation: end-to-end latency down 50 percent, traffic efficiency up 20 percent, and hard-braking events down 30 percent.^[1] These are the company's own comparisons between two generations of the system, and they describe aggregate behaviour rather than any single scenario. Latency matters because it sets how quickly the car can respond to something unexpected; hard-braking counts matter for ride comfort and for the risk of being hit from behind. "End-to-end" refers to going from sensor input to a driving decision without a chain of separately engineered stages in between.

Sensors and versions

ADS 4 introduces a high-precision solid-state lidar, an in-cabin lidar-vision sensor and distributed millimetre-wave radar.^[1] Lidar measures distance by bouncing laser light off objects, and a solid-state unit does this without a spinning mechanical assembly; radar does similar work with radio waves, which cope better with bad weather but resolve detail less sharply. The system ships in four versions, and the flagship, ADS Ultra, carries a dedicated highway L3 package.^[1] In the standard classification, L3 means conditional automation: within a defined operating domain the car performs the driving task and the human is not required to monitor continuously. Scoping the package to highways narrows that domain to motorway driving.

Where it sits

Qiankun is Huawei's route into the car industry: instead of building whole vehicles, it supplies the driver-assistance stack to established manufacturers, and the system's reach is measured by how many partner models carry it.^[2] The figure of 22 models given at launch marks how widely the line had already spread by the time the fourth-numbered generation arrived.^[2]

The launch also shows where the competition in intelligent driving has moved. The claims Huawei chose to lead with were not about a single dramatic capability but about latency, traffic flow and comfort — the measures that decide whether a driver-assistance system feels usable in everyday traffic — together with a simulated world used to manufacture the rare cases that ordinary road data supplies too slowly.^[1]