Robots Get a Productivity Score With New Worken Measure

A Productivity Score for robots took centre stage in Shanghai this week, as Chinese robotics firm AGIBOT unveiled a new way to measure how much useful work a machine actually delivers once it leaves the factory floor. The framework, called Worken, was introduced at the company's annual showcase, according to the announcement, and it aims to shift the industry's attention from how many robots are built to how effectively they perform real jobs, day after day, with little human help. For buyers weighing expensive automation, that distinction could soon matter more than any headline shipment figure.

The pitch is deliberately simple. A robot that rolls off a production line and sits idle, waits for instructions, or needs constant rescue by a human operator is not delivering the same value as one that works steadily through a shift, even if both count as a single unit sold. Worken tries to capture that gap by combining three ingredients: how many machines are actively doing tasks in real settings, how long they work on their own, and how their output compares with an established benchmark for the same job. The result is a single language for describing productive output across very different kinds of work.

AGIBOT's chief technology officer, Peng Zhihui, framed the idea as a response to a maturing industry. "Over the past few years, the robotics industry has made remarkable progress in hardware, AI models and large-scale manufacturing," he said at the event. "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?" The proposed measure, he argued, should reward machines that complete acceptable work autonomously and consistently, not merely machines that exist.

Counting useful work, not just units

The framework's three factors each solve a different blind spot in today's numbers. Counting only machines that are actively performing tasks, rather than units manufactured or held in storage, stops companies from inflating their progress with inventory. Measuring time spent working independently, rather than total time switched on, exposes systems that look busy while leaning heavily on human supervision. And judging output against a reference benchmark for a defined task, such as the acceptable work completed by a human under standard operating conditions, gives the number a meaning that raw hours alone cannot provide.

That last factor is also the hardest to standardise, because a successful factory operation and a successful public-service interaction produce different kinds of value. A machine tending a production line might be judged on cycle time, sustained operation, and successful task completion, while a service robot guiding visitors must be judged on speech understanding, responsiveness, and steady interaction with people. AGIBOT says the framework will be refined through real deployments and operating data, with task-specific benchmarks developed for each setting rather than a single universal yardstick.

The economics argument follows naturally. Hardware loses value over time, but software and AI models can improve, which means the same machine may deliver more useful work later in its life than it did at first. Measuring output alongside total lifecycle costs could help businesses judge whether a deployment is genuinely paying off or merely looking impressive in a demonstration video. For finance teams asked to approve fleets of costly machines, a credible Productivity Score could become as routine as fuel economy is for a vehicle fleet.

Hardware and models behind the pitch

Worken did not arrive alone. The showcase also introduced a new third-generation humanoid hardware platform that combines the foundations for movement, interaction, and task execution into a single architecture, with the goal of reducing integration complexity and improving responsiveness in real deployment sites. The platform adds a joint liquid-cooling system for continuous operation, protected joints for demanding environments, and a family of dexterous hands and grippers for fine manipulation. Each addition targets the same problem: robots that perform well in a laboratory often struggle with the heat, dust, and unpredictability of a working site.

On the software side, the company debuted a next-generation embodied foundation model built around a unified framework for reasoning, prediction, and action, designed to help robots anticipate the consequences of their actions and adapt to changing environments rather than repeating memorised sequences. A separate interaction model combines visual perception, speech, real-time dialogue, facial expressions, and physical actions into one continuous system, aiming for more natural communication in complex settings. The company also launched a heavy-duty industrial robot with twenty-one degrees of freedom and a dual-arm handling capacity of forty kilograms, built for demanding material-handling work in manufacturing and logistics.

Real deployments were presented as proof that the pitch is more than a slide deck. The company pointed to tablet production lines, semiconductor packaging and testing, machine tending, and warehouse operations, as well as visitor-service roles in a metro system and a tourism complex. In each case, the argument ran, deployment is not a one-off delivery milestone but an ongoing cycle of site integration, stable operation, and continuous improvement. That cycle is precisely what a Productivity Score is meant to track: not the day a robot arrives, but the months of work that follow.

Why buyers might care, and what to watch

Sceptics will note that Worken is, for now, one company's proposal rather than an independent standard, and the benchmarks that give it meaning have yet to be published in detail or tested by rivals. A measure designed by a manufacturer could favour its own machines, and rival firms may prefer their own yardsticks, leaving buyers to compare scores that were never computed the same way. The company acknowledges that the framework is evolving and says it hopes wider industry discussion will shape it into a shared reference, but that outcome is far from guaranteed.

Even so, the question Worken raises is one the whole industry will have to answer. Humanoid shipments have surged, and Chinese manufacturers now dominate unit volumes, yet most of those machines still work in entertainment, research, or data production rather than in factories. Investors and customers are beginning to ask what comes after the shipment chart, and a credible answer will need evidence from sustained operation rather than launch-day footage. A Productivity Score, if it gains independent backing, could separate serious deployments from expensive demonstrations.

The showcase also included the release of an operating system purpose-built for embodied intelligence, now being trialled by selected partners with plans to make it fully open source by the end of the year, alongside a donation to an open-source foundation. The ecosystem around the company already spans hundreds of universities, hundreds of enterprises, and a developer community numbering in the tens of thousands. Whether Worken becomes the industry's shared yardstick or remains a single firm's slogan, the shift it signals is real: robots are leaving the showroom, and from here on they will be judged on the work they actually do. Related coverage includes a rival humanoid production milestone, a major tech earnings story, and a record Canadian AI financing round.