XPeng bets on end-to-end VLA 2.0 to power its GX Robotaxi commercial rollout

bookwormengr · x · 2026-09-08

XPeng's self-driving stack is built on its proprietary end-to-end VLA 2.0 (Vision-Language-Action) model. Unlike rigid hand-coded driver-assist rules, it perceives, reasons, and drives like a human, working across cities, highways, and tight parking structures without expensive pre-mapped HD infrastructure. The same software foundation is driving the commercial rollout of its custom GX Robotaxi fleet into autonomous ride-hailing.

Original post →

More from Embodied

Embodied channel →