XPENG Demonstrates XNGP AI Driving to UN Regulators at WP.29 Session in Shanghai

XPENG placed international vehicle regulators into operational demonstrations of its XNGP ADAS, exposing live perception, motion planning, and control functions on urban and highway routes in Shanghai. Delegates from major markets—including EU members, Canada, Japan, the UK and the US—rode as observers, giving regulators direct telemetry and experiential data for evaluation.

The exercises highlighted an integrated safety stack that combines continuous driver status monitoring, adaptive human–machine interaction logic, and layered fail-safe controls. Observers noted system behavior across unpredictable traffic events and varied environments, providing real-world validation beyond simulation and closed-course testing.

XPENG also outlined its next-generation AI stack, labeled VLA 2.0 (Vision‑Language‑Action), which the company says reduces latency and information loss between visual inputs and vehicle actuation. That architecture frames the company’s Robotaxi roadmap and supports planned pilot operations scheduled for later this year in China.

By running live demonstrations during the UN WP.29 Informal Working Group on Automated Driving Systems session, XPENG narrowed the distance between regulator assessment and deployed systems. The forum’s technical focus on harmonized ADS rules means these demonstrations could inform global interoperability standards and design guidance for Driver Control Assistance Systems.

Technically, the session emphasized end‑to‑end system performance: stereo and monocular camera fusion, radar-assisted tracking, scene segmentation, policy-level decision modules, and closed-loop lateral and longitudinal control. Delegates recorded system responses to edge cases and how the stack transitions control between automated guidance and human operators.