---
title: "Unitree's Wang Xingxing: robot generalisation and manufacturing scale remain unsolved"
date: 2026-11-19
category: Humanoids & Robotics
site: NeuroAI
canonical: https://neuroai.site/a/na-auto-2026-11-19-unitree-generalisation-w
language: en
---

# Unitree's Wang Xingxing: robot generalisation and manufacturing scale remain unsolved

> At APEC, Unitree's Wang Xingxing said robot-model generalisation remains difficult and warned that manufacturing, reliability, service and compute must scale together.

Unitree founder Wang Xingxing used his APEC appearance to make an unfashionable argument: robot-model generalisation remains difficult, and the industry cannot scale on demos alone.

His point about manufacturing is the one investors underweight. Producing one humanoid that walks reliably on stage is an integration exercise; producing ten thousand that do not fail in customers' hands requires tolerances, supply chain depth and after-sales infrastructure that most robotics startups have never built.

Reliability compounds the problem. A robot with a 99 per cent success rate per attempt sounds strong until a task requires fifty consecutive successful attempts - which is roughly the requirement for a useful household or warehouse shift.

Compute is the third leg. Training generalist policies and running them onboard at low latency are different engineering problems, and both cost money that unit economics rarely cover at current prices.

His framing is useful precisely because Unitree is one of the companies best positioned to benefit from hype.

---

Published by NeuroAI (https://neuroai.site/) — https://neuroai.site/a/na-auto-2026-11-19-unitree-generalisation-w
Free to quote with attribution and a link to the original.
