Physix FrontierAutonomous driving in 2026: VLA, world models, and end-to-end—four paths, one destination
Autonomous Driving 2026: VLA, World Models, and End-to-End—Four Technical Routes, One Final Destination
In the spring of 2026, competition intensity in China's autonomous driving industry reached unprecedented levels. On March 2nd, XPeng announced the full rollout of its second-generation VLA, which He Xiaopeng called "opening the era of L4 physical world brains."
Four Technical Routes
- Tesla FSD V14: End-to-end neural network, making cars learning-capable, evolving intelligent agents
- XPeng VLA 2.0: Vision-Language-Action large model, leap in physical world understanding capabilities
- Huawei ADS 4.0: Continuous leadership in perception and planning/control dimensions, full coverage of urban NOA
- NIO/Li Auto: In-house chips + full-stack algorithms, differentiated routes
In 2026, the global L4-level Robotaxi industry officially entered a critical stage of scaled commercialization. Leading players' operational vehicle fleets reached thousands, and cities with active operations globally reached dozens.
Technical Consensus
End-to-end neural networks and world models have become industry consensus—"how cars understand the world" is becoming the core proposition of the industry. VLA, world models, and end-to-end approaches; a technical showdown about "how cars understand the world" is unfolding.
