Robots Atlas>ROBOTS ATLAS
Robotics & Hardware

Worken: AGIBOT Wants to Count Robot Work Hours, Not Units

Lady Robot16 November 2026 · 7 min read
Worken: AGIBOT Wants to Count Robot Work Hours, Not Units

At Innovation Day 2026 in Shanghai on 29 October, AGIBOT introduced Worken — a way of counting robots in hours of autonomous work rather than units shipped. The Chinese company used the same event to unveil the third generation of its AHA3 hardware architecture, the G2 Max industrial robot with a 40 kg dual-arm payload, and Link-U OS 1.0, an operating system it plans to fully open source by the end of 2026.

Key takeaways

  • Worken = number of active robots × autonomous working hours × output efficiency ratio
  • G2 Max: 21 degrees of freedom, dual-arm payload up to 40 kg, working height adjustable from 0.1 to 2.5 m
  • AHA3 adds joint liquid cooling and IP67-rated joints
  • Link-U OS 1.0 is running at selected partners, with full open sourcing announced for the end of 2026
  • The company's ecosystem spans 255 universities, 511 enterprises and more than 80,000 developers

Picture this situation

A phone factory buys fifty humanoids to feed parts to an assembly line. Six months later the supplier reports: fifty robots deployed, contract fulfilled.

Except twenty of them sit in a warehouse as spares. Ten are working, but an operator has to correct them every fifteen minutes. The remaining twenty genuinely do the job — just slower than the people they replaced.

The number in the report reads the same. The work actually delivered does not.

But here is the problem: how should this be counted so the number means anything?

That is the question AGIBOT is trying to answer with Worken. The figures in the scenario above are an editorial example, not deployment data.

What exactly does Worken mean?

The name blends “work” and “token”. The company translates it as “effective intelligent work hours”, and the formula itself is short:

Worken = number of active robots × autonomous working hours × output efficiency ratio

Each of the three factors closes off a different way of inflating the statistics.

The first counts only robots actually performing tasks in real environments — not units manufactured or held in inventory.

The second measures time spent working independently, not time the machine simply spends powered on. That is the difference between “the robot was energised for eight hours” and “the robot worked eight hours without intervention”.

The third compares output against a reference point for a specific task — for instance the amount of work a human completes in the same time under established operating standards.

Input
Deployment at a customer site
Worken
Count the active robots
Is the robot working autonomously?
YES
Hour countedAllow
NO
Hour discardedDeny
Worken
Compare output with the task benchmark
WorkenAllow

This is a simplified view of the proposed metric, not a description of a working settlement system.

Over the past few years, the robotics industry has made remarkable progress in hardware, AI models, and large-scale manufacturing. But as robots begin working in factories, warehouses, and service environments, we need to answer a more practical question: How much productive work are they actually delivering?

Peng Zhihui, co-founder, president and chief technology officer of AGIBOT

Why is counting deployed robots not enough?

Today the industry keeps score in deliveries. The company said last month that its 20,000th robot had rolled off the line, and IDC placed it first in global humanoid robot shipments for the first half of 2026.

The problem is that a delivery says nothing about what happens after deployment.

Two deployments with the same number of machines can deliver entirely different productivity — depending on how independently the robots operate, how often they need a human, and how much acceptable work they actually complete.

There is a second, purely accounting side. Hardware depreciates, but better models can let the same machine perform more useful work over time. Without an output metric there is no way to set that against lifecycle cost, and that is exactly the number an operations director weighing a purchase cares about.

Worken is not an industry standard. It is one company's proposal, and that company admits the framework will be refined on data from further deployments.

What else did AGIBOT show besides the metric?

The hardware foundation is AHA3 — a third-generation architecture described as “three intelligences in one”: locomotion, interaction and manipulation in a single system. New additions include joint liquid cooling, designed for continuous industrial operation, and IP67-rated joints. The whole thing sits on integrated compute and Wi-Fi 6 plus 5G connectivity.

G2 Max also debuted — a heavy-duty industrial robot with 21 degrees of freedom, a dual-arm payload up to 40 kg and a working height adjustable from 0.1 to 2.5 m. Force-controlled manipulation and omnidirectional mobility come with it.

The gripper family grew by three OmniHand 3 variants plus the tactile OmniPicker 3 gripper. Separately the company showed its GO-3 Preview foundation model, which we covered earlier.

The last piece is Link-U OS 1.0 — an operating system the maker calls “the industry's first purpose-built for embodied intelligence”. It sits on a unified hardware abstraction layer and connects hardware with applications. The project was donated to the OpenAtom Foundation, alongside a pledge to jointly build an open-source community for humanoid robots and unify technical interfaces.

What do these names mean?

Four acronyms recur most often in this announcement:

  • Worken — a metric for work performed by robots. The equivalent of payroll work hours, only calculated for machines and adjusted for how much of that hour was genuinely autonomous.
  • AHA3 — AGIBOT Humanoid Architecture, third generation. A shared hardware base for the company's different robots instead of a separate build per model.
  • Link-U OS — the robot's operating system. Its role resembles Android on a phone: it hides hardware differences so an application need not know which motor sits in a joint.
  • WRAM — World Reasoning Action Model, the framework behind GO-3 Preview. It is meant to let a robot anticipate the consequence of a motion before performing it.

Together these are an attempt to close the loop: shared hardware, a shared system and one number for settling output.

Is it working yet?

Partly — and each piece at a different stage. Worken is a measurement framework proposed on 29 October 2026, not an adopted standard. G2 Max and the new grippers are product launches with no published prices or delivery dates. Link-U OS 1.0 has, according to the company, been validated on mass-produced robots and is being trialled by selected ecosystem partners, with full open sourcing announced for the end of 2026.

G2 family deployments are real and named. At computer maker Longcheer the robots feed material on a production line, with results measured by cycle time, sustained operation and task completion. The company also cited Guangzhou Metro and the Chimelong tourism complex, where machines guide visitors and answer questions.

Why it matters

For three years the humanoid industry has raced on deliveries and spectacular demos. Worken shifts the argument to something uncomfortable for every supplier: how many of those machines really work without a minder. A metric invented by a manufacturer will always be suspect, but the mere fact that the shipment leader proposes being judged on output rather than volume changes the language of customer negotiations.

What's next

  • Full open sourcing of Link-U OS 1.0 announced for the end of 2026 — until then the system stays with selected partners
  • The company said it will refine Worken on deployment data and build task-specific benchmarks
  • The Link-U OS project donated to the OpenAtom Foundation is meant to anchor a shared open-source community for humanoid robots with unified technical interfaces

Sources

Share this article