--- title: "UBTECH's Walker S humanoids are now working real shifts inside Chinese auto plants" date: 2026-09-29 category: Humanoids & Robotics site: NeuroAI canonical: https://neuroai.site/a/na-robot-ubtech-walker-s language: en --- # UBTECH's Walker S humanoids are now working real shifts inside Chinese auto plants > UBTECH's Walker S industrial humanoid (人形机器人) has moved from trade-show dancing to paid, repetitive work on Chinese factory floors. Twenty units went into a Dongfeng Liuzhou motor plant in 2025, and the company passed 1,000 units built by November. The interesting question is no longer whether a humanoid can walk — it is whether a cluster of them can out-earn their cost. On a Zeekr assembly line in Ningbo, a 172-centimetre silver humanoid (人形机器人) sorts parts between stations and checks badges on car bodies. It is not a prototype on a stage, and it is not a research mascot. It is on the floor because someone signed a purchase order. UBTECH (优必选) has spent more than a decade building humanoids, and the past two years are when the story stopped being about choreography and started being about throughput. ## From Spring Festival to shop floor The company's robots first became famous for dancing at the 2016 Spring Festival Gala. The pivot to industry came with the **Walker S** line, launched in 2023 for industrial scenes. By early 2025, **20 Walker S1 units** were slated for deployment at **Dongfeng Liuzhou Motor (东风柳汽)** in the first half of the year — described at the time as the world's first batch of humanoids moving into an automobile factory at scale. Those were not the only floors. Walker S-series units have been in training or deployment at **Geely (吉利)**, **BYD (比亚迪)**, **Foxconn (富士康)** and **SF Express (顺丰)**, learning the three jobs UBTECH now treats as its wedge: **sorting, handling and quality inspection**. ## The thousand-unit milestone In November 2025, UBTECH rolled out its **1,000th Walker S2** — the third-generation model that can swap its own battery in about three minutes, so it never has to clock out to recharge. That number is the real signal: this is a production line making production-line workers, not a craft workshop. The business behind it: - **2025 revenue passed 2 billion RMB** (about **US$281 million**, roughly **HK$2.19 billion**), up **53.3%** year on year. - Full-size embodied humanoid revenue jumped **2,203.7%** to **8.2 billion RMB** (about **US$115 million**, roughly **HK$898 million**), on **1,079 units sold** — a global lead the company cites. - UBTECH delivered **over 500 units** in 2025 against capacity above **1,000**, and plans to lift annual capacity to a **10,000-unit scale in 2026**. - Its single largest order reached **2.64 billion RMB** (about **US$37 million**, roughly **HK$289 million**), reported as a record for a humanoid-robot deal in China. ## Why clusters, not lone machines The founder, **Zhou Jian (周剑)**, has been blunt that a single humanoid is not the product — a coordinated squad is. UBTECH's "group intelligence" stack, including a **BrainNet** software architecture and a self-developed embodied large model (大模型) called Thinker, lets multiple units divide tasks, avoid each other and recover from surprises without a human in the loop. The economics are the point. UBTECH says core-component localization is already **above 90%**, and it targets a mass-production cost of **under US$20,000 per unit** once it reaches the 10,000-unit scale — a figure that lines up with what other global humanoid makers cite as the threshold for factory adoption. ## What is still unsolved Honesty matters here. A humanoid that sorts boxes is not yet a humanoid that replaces a worker. UBTECH itself reports that, on handling tasks, a single unit's efficiency versus a human rose from about **30% to roughly 45%** early in 2025, with a target above **60%** as group intelligence matures. The machines currently shine at the structured, repetitive, high-turnover stations — exactly where factories say they cannot hire enough people. And the home robot is a different mountain. Zhou has repeatedly said humanoids in everyday homes are a **5-to-10-year** problem, not a next-year one. ## What readers can do now - **If you run a factory**, pilot Walker S-class humanoids on the three proven stations — handling, sorting, inspection — where the payback math already works, and demand cluster features, not single-unit novelty. - **If you follow robotics supply chains**, track the >90% localization rate and the US$20,000-at-scale cost target; those two numbers, not demo videos, decide whether humanoids become line items or science projects. - **If you report on the sector**, separate "units in training" from "units bought and deployed" — UBTECH's own 500-plus deliveries against 1,000-plus capacity is the metric that separates reality from roadmap. ## Honest limitations Quantitative facts (20 Walker S1 units to Dongfeng Liuzhou in H1 2025; training at Geely, BYD, Foxconn, SF Express; 1,000th Walker S2 in November 2025; 2025 revenue 2 billion RMB / +53.3%; full-size humanoid revenue 8.2 billion RMB / +2,203.7% / 1,079 units; 500+ delivered, 1,000+ capacity, 10,000-unit 2026 plan; 2.64 billion RMB largest order; >90% localization;