Industry News
AGIBOT Proposes Worken to Measure Autonomous Robot Work
Photo: brewery robot arm (2024), not AGIBOT. Worken proposes a way to account for autonomous robot work. Its usefulness depends on task definitions, quality thresholds and clear human-assistance rules.
AGIBOT has proposed Worken, a measure of productive robot work, at its October 29 Innovation Day in Shanghai. The company combines active robot numbers, autonomous working time and task output into one calculation. It is a measurement proposal, with no independently validated Worken fleet result accompanying the announcement.
For buyers of humanoid robots, the useful question is what a machine contributes during a shift. A delivery count cannot answer that. Worken gives the discussion a compact formula; making its results comparable will require clear rules about tasks, quality and human assistance.
Key Stats: illustrative inputs for a uniform cohort
N robots
Active task-performing cohort
H hours
Autonomous work per robot
E ratio
Output relative to a task benchmark
Symbols explain the calculation. They are not reported fleet measurements.
What Worken would measure
In AGIBOT's announcement, the count excludes machines sitting in inventory, the clock measures autonomous work rather than powered-on time, and output is compared with a defined task benchmark. One possible reference is acceptable human output under an established operating standard. The company describes the framework as evolving.
Those distinctions matter even before a score is calculated. Ten robots listed on a purchase order and ten robots contributing to a production line describe different conditions. Likewise, a robot that remains switched on through a stoppage has accumulated time without necessarily completing useful work. A result needs enough detail to distinguish those cases.
The word “acceptable” also carries weight. Counting every grasp would reward a machine for moving a defective component as readily as a good one. An operator evaluating an assembly task would need a pass condition tied to the actual process: the correct part, in the correct location, meeting the relevant quality requirement. That is an evaluation requirement, not a claim that AGIBOT has published a complete audit protocol.
An illustrative shift, not a reported result
Consider a hypothetical group of ten identical robots doing the same task. Each performs six hours of autonomous work in an eight-hour shift. Assume its accepted output per autonomous hour equals 80% of the chosen reference rate. Multiplying 10 × 6 × 0.8 gives 48 benchmark-equivalent work hours.
That arithmetic does not establish that 48 hours of paid human labor disappeared. It does not describe a measured AGIBOT installation. It shows how the inputs interact: improve the accepted output rate while holding the other assumptions fixed, and the total rises. Keep machines powered on for longer without increasing autonomous task time, and that alone adds nothing.
For a mixed fleet, applying one common duration and efficiency ratio could obscure large differences. An auditable implementation could calculate contributions separately for comparable tasks and operating periods before adding them. It would also need to explain any averaging. Otherwise, a strong machine on an easy task could mask a weak one on a difficult assignment.
How it compares with familiar factory measures
Manufacturing already has ways to connect time with useful output. Equipment-monitoring supplier Vorne explains Overall Equipment Effectiveness, or OEE, as availability multiplied by performance and quality, expressed as a percentage. It describes the efficiency of a machine, line or cell. Worken's proposed combination addresses a different reporting question: a fleet's autonomous work contribution over a period.
| Indicator | What it describes | What the number alone leaves open |
|---|---|---|
| Units delivered | Hardware supplied | Whether it is working or awaiting integration |
| Autonomous task hours | Time recorded under the chosen autonomy rule | Speed, accepted output and task difficulty |
| OEE percentage | Availability, performance and quality | The amount of human assistance behind production |
| Proposed Worken total | Active cohort, autonomous duration and relative output | Full cost, safety and comparability across benchmarks |
This comparison is an interpretation of what the measures answer, not evidence that one has replaced another. A factory could reasonably retain both equipment-loss tracking and an autonomy-specific account. If their totals disagree, the useful response would be to inspect the clocks, denominators and quality rules behind them.
Autonomy needs a setting
The importance of context predates the current robot boom. The U.S. National Institute of Standards and Technology's ALFUS research framework characterized autonomy through mission complexity, human independence and environmental difficulty. It was a framework for unmanned systems, not an assessment of Worken.
That distinction offers a practical reading rule. An hour sorting identical objects in a fixed tray and an hour navigating a changing public space are not interchangeable evidence of capability. Reporting the setting alongside a total would make the result more useful. Removing the setting could make apparent progress reflect an easier assignment instead of a better robot.
AGIBOT's wider platform push
The event also presented AHA3 hardware and GO-3 Preview. Link-U OS 1.0 is being trialled with selected partners, with full open sourcing planned for the end of 2026, according to the company. The company-issued release supplies the announcement's date and scope.
AGIBOT also publishes an AgiBotWorld2026 dataset card, describing real-world G2 training data across commercial, home and general settings. Training examples help describe how a model is developed. They do not, by themselves, measure the productive output of an operating customer fleet.
For AGIBOT, a productivity measure could become a way to assess the entire system rather than an isolated model. That is an analytical possibility, not demonstrated performance. A software update could improve task completion yet create more recovery work elsewhere. The relevant comparison would hold the task and acceptance conditions steady and examine the whole shift.
What buyers would need beside a score
A useful operating record would identify the measured task, period, robot cohort and output standard. It would show failed attempts, rejected work and human interventions as well as successful cycles. The following are evaluation questions raised by the proposal, rather than requirements attributed to an unpublished Worken specification.
Where does assistance enter the clock? A technician might reset a fault, clear a fixture, send a remote command or prepare the next batch. Each action could change the amount of labor needed around a robot. A report should explain how those intervals are treated and how much supporting human time the operation consumes.
The distribution of interruptions matters as well as their sum. Six productive hours interrupted once near a scheduled break could fit a process differently from six hours broken into dozens of short recoveries. Both might produce the same total, while demanding very different attention from a supervisor. A shift log that records interruptions would preserve information that an aggregate necessarily loses.
Does the comparison survive a task change? Switching from an awkward object to an easy one can raise output without improving the underlying system. Keeping task definitions and benchmark versions visible would let an operator distinguish a software improvement from a change in the assignment. It would also discourage combining unlike activities into an unexplained league table.
What does the work cost? Hardware payments, integration, maintenance, supervision and energy still belong in a business case. A productive-hours total cannot supply those missing prices. Nor does it establish a safe working arrangement. Buyers need cost and safety evidence alongside throughput, rather than treating one scalar as a purchasing verdict.
What to watch next
The next useful milestone would be a reproducible measurement definition accompanied by operating records. A before-and-after software comparison on a fixed task would be more informative than an isolated score. An independent operator reporting the same fields would make the proposal easier to assess beyond the vendor's own examples.
Publication of benchmark versions, intervention rules and quality thresholds would also allow buyers to ask narrower questions. They could compare a robot against their own process instead of accepting an industry-wide ranking. The essential test is whether another evaluator can reconstruct the result from the underlying work record.
The reporting window should be explicit, too. A selected hour after setup answers a narrower question than a full week including changeovers and maintenance. Showing both can help separate peak capability from the sustained contribution a customer can plan around.
Frequently asked questions
What is Worken?
It is AGIBOT's proposed measure of productive robot work. Its calculation combines active robots, autonomous working time and output relative to a task benchmark.
Is the 48-hour example an AGIBOT result?
No. It is a hypothetical calculation using ten robots, six autonomous hours each and an output ratio of 0.8. It explains the arithmetic and is not a measured fleet result.
Is Worken already an industry standard?
The announcement does not establish that status. A vendor proposal needs published definitions and evidence of adoption before readers can treat scores as comparable across suppliers.
Does a productive-hours score show jobs replaced or money saved?
No. Those conclusions require separate evidence about staffing, supporting labor and total costs. A normalized work total does not directly translate into either headcount or profit.
A measurement proposal with a practical test
A simple formula can improve the questions asked of robot suppliers. Its value depends on whether the underlying record is specific enough for a customer to check. For Worken, task definitions and transparent accounting will determine whether a headline number becomes useful operating evidence.
The takeaway: Evaluate a productivity score together with its task, quality standard, autonomy rules and supporting labor.
Hero photograph: robot arm on Spoetzl Brewery's keg-handling line in Shiner, Texas, photographed June 1, 2024. Industrial-work context; this is not AGIBOT equipment or a Worken measurement. © 2024 Larry D. Moore / Wikimedia Commons, CC BY 4.0. Wikimedia resized rendition; responsive display may crop. No color alteration or endorsement is implied.