L4 Autonomous Driving Starts to Do the Math

Gasgoo Munich- On the tarmac at Urumqi Tianshan International Airport, autonomous vehicles are no longer a novelty.

Like other ground support equipment, they shuttle between aircraft stands, hauling cargo, ferrying luggage, and transporting crew. Once a flight lands, the system dispatches tasks and the vehicles follow instructions to their designated workstations. Far from a technology demonstration, these autonomous vehicles function more like standard pieces of the airport's operational machinery.

That very ordinariness is the most significant shift for Level 4 autonomy.

Image Source: Gasgoo

Five years ago, Xinjiang Airport Group began partnering with UISEE. The first batch of vehicles arrived in 2021, and by late 2024, autonomous tractors had entered commercial operation in the airport's core tarmac areas. Today, Tianshan Airport has deployed more than 70 autonomous vehicles, including tractors and buses. Among them, 52 autonomous tractors handle over 90% of cargo and mail transfer duties, accumulating over 1.6 million kilometers of autonomous operations.

Yet another set of figures disclosed by the airport authority deserves more attention than that 1.6 million kilometer mark: operating costs for traditional diesel tractors run about 2.2 yuan per kilometer, but with new energy autonomous vehicles, that cost drops to roughly 0.8 yuan.

This suggests that once autonomous driving is truly integrated into production workflows, how the industry evaluates Level 4 is shifting.

For the past decade, the industry focused on whether vehicles could drive themselves, how often they required human intervention, and who could be the first to remove safety drivers. Now, the questions are different: How much work can it do in a day? How much cost does it save? How many people are needed to maintain it? And can that capability be replicated at the next airport?

Level 4 is gradually evolving from a race over technical capability into a test of production efficiency and business models.

Why Airports?

To balance the books, we first need to answer a question: Why is Level 4 achieving scale operations in scenarios like airports first?

After all, for a long time, the biggest potential for Level 4 autonomy was seen in Robotaxis.

Open urban roads present a massive array of long-tail scenarios; conquering them implies a high technological ceiling. Taxis and ride-hailing services represent a huge market where the commercial value of removing the driver is easiest to understand.

Therefore, even as capital has retreated, companies like Baidu Apollo Go, Pony.ai, and WeRide continue to expand their Robotaxi operations.

Yet over the past few years, another path for Level 4 commercialization has gradually become clear.

Scenarios like airports, ports, mines, and industrial parks may lack the massive market potential of Robotaxis, but they share a common characteristic: relatively clear operational boundaries, where the safety, efficiency, and cost benefits of autonomy are easier to quantify.

Wu Gansha, co-founder and CEO of UISEE, used a vivid metaphor to explain why the company chose these scenarios early on: "The initial autonomous driving system is like a golden shovel—it's expensive, so you have to dig in rich mines first."

Image Source: Gasgoo

That statement actually captures the universal dilemma facing early Level 4 commercialization.

Sensors, computing platforms, R&D, and engineering deployment all come with high costs. If you enter a scenario with low labor costs and low vehicle utilization, the economics won't add up—even if the technology enables autonomous driving.

Consequently, different Level 4 players are searching for their own "rich mines": Robotaxis bet on high-frequency urban mobility, mines target high-risk and high-intensity roles, and ports seek high-frequency repetitive transport, while airports face a different set of demands.

Take Tianshan Airport: flight support involves not only a large volume of repetitive transport tasks but also extreme weather like bitter cold, blizzards, and sandstorms. Support efficiency during peak hours, safety pressures, and the need for round-the-clock operations together form the commercial foundation for autonomous driving in airports.

The cost difference mentioned earlier—2.2 yuan versus 0.8 yuan—makes that value concrete for the first time.

Of course, this data cannot simply be equated to "autonomous driving being 64% cheaper than manual driving." There are inherent energy cost differences between diesel and new energy vehicles, and autonomous systems involve vehicles, sensors, software, depreciation, and maintenance. A true ROI judgment requires a more complete total cost of ownership analysis.

But at the very least, Level 4 is now being judged on production efficiency rather than purely technical metrics.

According to a Frost & Sullivan report, calculated by 2025 revenue, UISEE holds a 90.5% share of the Level 4 commercial autonomous driving market in airport scenarios across Greater China. As of May 2026, it has cumulatively deployed over 1,400 Level 4 vehicles and kits, covering 6 countries and regions and 249 customers.

However, that 90.5% figure also needs to be understood within the context of the market's development stage. Wu Gansha estimates that the overall penetration rate of autonomous driving in airport vehicles remains in the low single digits.

The coexistence of high market share and low penetration rates precisely indicates that airport Level 4 autonomy is still in the early stages of commercialization.

The Real Challenges Begin Once You're Running

Finding a "rich mine" doesn't guarantee a viable business model.

The biggest difference between airports and public roads is that once an autonomous vehicle enters the flight support workflow, it is no longer a test car proving technical capability—it is a piece of production equipment that cannot easily stop working.

Wu Gansha summarized this requirement: "If autonomous driving scores 99 out of 100, in a sense, it's still a zero. If there's a serious problem just one day out of 100, customers will struggle to truly use the product."

This means vehicles must not only know how to drive but must do so stably over the long term.

Tianshan Airport happens to amplify this requirement.

Image Source: Gasgoo

Extreme cold, blizzards, sandstorms, high heat, and low visibility—every one of these extreme conditions can affect sensors, vehicle control, and even electrical systems. More troublesome is that some extreme weather data cannot be fully replicated in a lab. Wu Gansha noted that a problem unsolved one winter sometimes requires waiting for the same weather the following year to continue verification.

But once truly integrated into airport operations, the challenges extend beyond the vehicles themselves.

"After a flight arrives, electronic instructions are sent directly by the system specifying how many vehicles are needed and where and when they should arrive at the stand, eliminating reliance on verbal communication."

Wu Gansha revealed that a significant part of the early work on the Tianshan Airport project involved integrating the autonomous system into the airport's existing dispatch and support infrastructure.

This also explains why the on-site presentation focused less on end-to-end technology, VLA, or world models, and more on seven-layer safety systems, weak-network operation, vehicle health analysis, and predictive maintenance.

Image Source: Gasgoo

In the OAM (Operations and Maintenance) system demonstrated by UISEE, incident response is required at the second level, remote recovery within minutes, and on-site rescue within 10 minutes, alongside daily health analysis and predictive maintenance.

Looking further ahead, automatic hitching and unhitching, automatic charging, and automatic loading and unloading have become part of product evolution. The reason is simple: if a vehicle no longer needs a driver but still requires a large staff to hitch, charge, and load, the economic value of "full-process autonomy" is diminished.

Therefore, once Level 4 enters real production scenarios, competition actually expands from vehicle intelligence to the entire operational system.

Algorithms determine whether a car can drive itself; engineering capability determines whether an unmanned fleet can work stably.

Now that Tianshan Airport has used 1.6 million kilometers to prove that autonomous vehicles can participate in airport support over the long term, the next question naturally becomes: Can this capability be replicated elsewhere?

Success at One Airport Doesn't Mean Commercial Success

The autonomous driving industry does not lack successful projects; what is truly scarce are successful projects that can be replicated at low cost.

This is the next critical threshold for airport Level 4 autonomy.

Judging by the project timeline disclosed by UISEE, from early exploration at Hong Kong International Airport to Singapore's Changi Airport and then to Tianshan Airport, the deployment cycle has shortened significantly.

Specifically at Tianshan Airport, building on previous experience, the subsequent deployment and delivery of 40 autonomous tractors took less than a month.

But making it work at one airport is not simply equivalent to achieving standardized replication.

Different airports have variations in climate, road environment, aircraft types, operational workflows, and management rules. If entering a new airport requires re-collecting data, adjusting systems, and deploying large numbers of engineers, then more projects mean more people, making it difficult to achieve true economies of scale.

Wu Gansha addressed this issue himself: "In the future, if there are 100,000 AI drivers, they might be distributed across 10,000 locations. If we have to send engineers to stay on-site for two months at every location, this model clearly cannot scale."

So UISEE has set a goal of "out-of-the-box" deployment.

But to achieve "out-of-the-box" readiness, it is not enough for a single company to improve technology reuse rates; industry standards must also catch up.

According to an individual involved in on-site standards research, the study of standards for airport autonomous vehicles has resumed this year. The current path being considered involves "leading with group standards, then converting them to industry or national standards once mature."

The currently planned standard system is divided into three layers: general basics, product application, and operations management. Over 50% of basic requirements plan to adopt existing national intelligent connected vehicle standards, while supplementing airport-specific requirements. For test scenarios, 42 general scenarios are retained, 28 airport-specific scenarios are added, and 32 general clauses inapplicable to airports are removed.

Standards essentially solve the replication problem: what kind of vehicles can enter airports, what tests they must pass, how they are accepted, and how they should be operated and maintained.

In the past, many projects relied on "case-by-case" negotiations between companies and airports. Only after standards are gradually unified can products truly possess the foundation for cross-airport replication.

Next, Xinjiang Airport Group plans to expand autonomous driving services from Urumqi to all 27 operating airports in Xinjiang.

Compared to "1 million AI drivers," these 27 airports might serve as a more realistic metric for observation.

If a system already proven at Tianshan Airport can enter a second, fifth, or twenty-seventh airport with continuously declining delivery times, labor input, and costs, then Level 4 will truly have begun the transition from project to product.

Before 1 Million AI Drivers, Do the Math First

Replicability must ultimately withstand the test of the business model as well.

UISEE's next step is to attempt a shift from "selling vehicles + maintenance" to a continuously billable "AI driver" model.

According to its plan, future business models will include vehicle leasing, retrofit autonomous driving kits, pre-installed autonomous-ready models, and charging by service cycle or mileage. The core logic is to reduce upfront customer investment and turn autonomous driving from a one-time project into a continuous service.

Wu Gansha even offered a bold long-term vision:

"In the next 5 to 10 years, if we can deploy 1 million AI drivers and charge a service fee of $10,000 per AI driver annually, that corresponds to $10 billion in subscription revenue."

The potential of this story is vast, but it is first and foremost a corporate long-term vision, not a performance forecast.

The reality is that as of May 2026, UISEE's cumulative deployment scale remains just over 1,400 Level 4 vehicles and kits.

From 1,400 to 100,000, and then to 1 million, the gap to be bridged involves more than just vehicle numbers.

Image Source: Gasgoo

Hardware costs must fall further, delivery cannot rely long-term on engineers stationed on-site, and operations need further automation. Furthermore, as the business moves from airports to buses, sanitation, heavy trucks, and urban delivery, how much of the accumulated data and technology can be reused will also need re-verification.

The same applies to overseas markets. Different countries and regions have varying regulations, certifications, infrastructure, and operational systems—"making it work at a Chinese airport" does not naturally equate to "replicability at global airports."

So, standing at Tianshan Airport and looking back at that 1.6 million kilometers, the significance of this number is not just proving that autonomous driving can travel far enough.

It is more like a dividing line in the commercialization of Level 4 autonomy.

After hitting 1.6 million kilometers, the real question is no longer whether it can reach 2 million or 3 million kilometers next, but whether this system can enter the next airport at a lower cost and replicate the same efficiency and economic value there.

Ten years ago, autonomous driving companies first had to answer: Can the car drive itself?

Today, the question is becoming: Can a driverless car be more reliable and more economical than a manned one, and can it be replicated at a low enough cost to the next airport, the next city, and the next country?

For UISEE, the distance from over 1,400 Level 4 vehicles and kits to its envisioned 1 million AI drivers is still long.

The same is true for the entire Level 4 industry.

The real competition in the next stage of autonomous driving may no longer be about who drives the car more like a human, but who can first turn "unmanned" into a standardized, replicable, and economically viable production tool.