March 31 marked a milestone for China's intelligent connected vehicle industry. Gasgoo has learned that Changan Automobile has officially obtained an L4-level Robotaxi test license. The license was jointly approved by six departments in the Chongqing Yongchuan Policy Pilot Zone, signifying that Changan can conduct autonomous driving tests in the region in all scenarios and without human intervention.
The two most striking keywords of this license are "full compliance" and "all scenarios." The former means passing the pilot zone's most stringent joint review, while the latter represents the ability to operate without a safety driver in designated areas.
In spring 2026, when L3 autonomous driving was just entering pilot admission and the industry was still debating "whether to skip L3," Changan answered with this L4 license: no leapfrogging, but parallel advancement.
The Path to Large-Scale Autonomous Driving Production
This L4 Robotaxi is not an isolated prototype. It is built on Changan's self-developed Tianshu end-to-end large model, sharing the same technical foundation as its mass-produced L2 and L3 models, with a high degree of reuse of the underlying algorithms.
This means that Changan's high-level autonomous driving is not confined to laboratory vehicles. It already possesses the ability to migrate from driver assistance to high-level autonomous driving at scale.
In 2025, Changan Automobile officially released its self-developed "Tianshu Intelligent Driving" system, an end-to-end interactive navigation assistant. The system integrates core technologies in driving, cockpit, and chassis, including end-to-end navigation, multimodal AI models, distributed electric drive, and steer-by-wire chassis. It achieves four-wheel independent control, reduces drive and brake response time to 10 milliseconds, lowers hydroplaning risk by 76%, and can correct the trajectory within 0.1 seconds in the event of a tire blowout.
Changan plans to further enhance perception capabilities. By the third quarter of 2026, the company expects to increase the precision of its self-developed satellite-architecture LiDAR to more than 200 lines, triple computing power, while reducing costs by 30% compared with peers. Once fully matched with the Tianshu end-to-end model, these upgrades will significantly improve recognition capabilities in extreme scenarios such as night driving, car following, and congestion.
Chongqing: The Litmus Test for the L4 License
The real weight of this L4 license lies in the test site — Chongqing.
Chongqing is renowned for its "8D magical road conditions," with stacked overpasses, complex tunnels, and steep sharp curves, coupled with frequent fog. It is called the "ultimate examination room" for the autonomous driving industry.
Such a demanding testing environment builds a unique "terrain barrier" for the reliability of Changan's L4 solution. If it can work in Chongqing, it can adapt to anywhere.
From L3 to L4: Balancing Technology Rhythm and Industry Landscape
Obtaining the L4 license is not an isolated event, but the latest move in Changan's high-level autonomous driving strategy.
Just on December 20 last year, Changan obtained the country's first L3-level autonomous driving license plate, "渝AD0001Z". From L3 "compliant road use" to L4 "driverless" testing, Changan took only three months.
This dual-track rhythm reflects the deep divergence in the current industry route.From late 2025 to early 2026, L3 pilot models from Changan and BAIC Arcfox successively obtained MIIT market access and began road testing on designated routes. Meanwhile, there is also a "leapfrog camp" urging the industry to skip L3 entirely and go straight to full-scenario L4 autonomous driving.
Beneath the surface clash of routes lies a rational calculation at the commercialization tipping point. L3 resolves the liability and insurance issues of human-machine co-driving, clearing the regulatory path; L4, by contrast, aims to ultimately eliminate the safety driver and seize the commanding heights of future mobility services.
Changan's strategy is more like a "dual-track system": it accumulates real-world data and user feedback through mass-produced L3 models, while using Robotaxi permits to conduct future-oriented L4 validation. This approach avoids both the risk of a pure-L3 route getting stuck in a "transition period" and the pitfall of a pure-L4 route whose commercialization is too far off.
The key lies in "shared technical origins." L2, L3, and L4 all rest on the same underlying algorithmic architecture. Advances in higher-level capabilities feed back into the assisted-driving experience of mass-produced vehicles, while data from those vehicles continuously trains higher-level models, forming a virtuous cycle.