UISEE TECH announced on the evening of August 19 that its fleet of Level 4 true driverless autonomous vehicles has surpassed 1.6 million kilometers of cumulative operational mileage at Tianshan International Airport, with total true driverless operations across all airports exceeding 6 million kilometers, all achieved with zero at-fault incidents.

According to Wu Gansha, co-founder, chairman and CEO of UISEE TECH, nearly 100 of the company's unmanned vehicles are currently shuttling across the tarmac at Urumqi Tianshan International Airport. Among them, 52 driverless tow tractors handle more than 90% of the airport's cargo and mail transport, with a peak daily volume of 800 trailers. Additionally, 20 driverless shuttle buses operate across 13 stops, covering office areas, access points and the apron.

"The 1.6 million kilometers were accumulated through real-world operations of the entire fleet, which is far more meaningful than single-vehicle mileage," Wu noted, adding that these figures must be viewed within a far more demanding framework: at an airport, a safety score of 99.9 is equivalent to zero.

In Wu's view, civil aviation represents the best application scenario for physical AI, yet the flip side of being "best" is being "most difficult." The apron has no lane markings, satellite positioning is severely disrupted by aircraft takeoffs and landings, and operations must endure blizzards at minus 20 degrees Celsius as well as surface temperatures reaching 60 to 70 degrees in summer. Any positioning drift could potentially send a vehicle onto a taxiway.

"Mistakes in the digital space can be redone, but in the physical world, a small error could lead to a catastrophic disaster," Wu explained. He emphasized that in UISEE TECH's product framework, safety always comes first, describing it as "the '1' that precedes all the zeros."

To meet civil aviation's most stringent safety requirements, UISEE TECH has built a seven-layer progressive safety architecture. Each vehicle generates hundreds of states, which through interactions produce tens of thousands of state transitions. The foundational safety layer ensures any anomaly can be detected and halted. Above that sits a redundant design with at least six positioning mechanisms that back each other up. Further up, leveraging the industry's largest airport-specific database, algorithms are evolving from rule-based approaches toward VLM, VLA, large models and world models, with reinforcement learning employed to train increasingly safer algorithms.