China Robotics - Physical AI: From NeuralAxis blueprint to Unitree

研报英文原文证据摘录

China Robotics - Physical AI: From NeuralAxis blueprint to Unitree

Global Markets Research

28 June 2026China Robotics

EQUITY: TECHNOLOGY

Physical AI: From NeuralAxis blueprint to Unitree Research Analysts

Advanced Manufacturing

Blueprint to product: Unitree WVLA2.0 operationalises the Frank Fan - NIHK

frank.fan@nomura.comNeuralAxis doctrine

+852 2252 2195

On 15 June 2026 we visited Unitree (Unlisted) to update on the company's latest

Donnie Teng - NIHKdevelopments (TakeawaysfromUnitreevisitaheadofitslisting). This report frames donnie.teng@nomura.com

Physical AI commercialisation around two complementary layers: a "system-architecture +852 2252 1439

blueprint and safety doctrine" for landing Physical AI in the real world, and, following the

test release of Unitree's WVLA2.0 (World-model Vision-Language-Action) embodied large

model, our read on how Unitree turns that blueprint into a shipping product and a

commercial roadmap through model fusion and hardware-software co-design.

NeuralAxis by NXP: The reflex-first blueprint for physical AI

The NeuralAxis (Neural Axis Architecture) framework was proposed by NXP (NXPI US,

Not rated), unveiled by President and CEO Rafael Sotomayor at the COMPUTEX 2026

keynote. Its premise is that Physical AI's binding constraint is not scaling language-model

reasoning but engineering a low-latency reflex layer akin to human unconscious

response, consistent with Moravec's paradox. NeuralAxis mirrors the human nervous

system across three decoupled yet coordinated tiers: a reasoning layer (cortex, ~300ms),

a coordination layer (cerebellum) for motion control and balance, and a reflex layer (spinal

cord, as low as 40ms) pushed to the edge near actuators—jointly delivering low latency,

distributed control and high energy efficiency.