AGIBOT proposes ‘Worken’ metric to measure robots’ real-world work

On October 29, 2026, at AGIBOT Innovation Day 2026 in Shanghai, AGIBOT proposed Worken, a framework to measure the amount of useful work robots perform in real-world settings. Its formula is “number of operating robots × autonomous operating time × output efficiency ratio.” Rather than focusing only on the number of robots manufactured or deployed, it evaluates the work they actually perform.

AGIBOT’s proposal adds autonomous operating time and the amount of work that meets quality standards to evaluations that are difficult to make using deployment numbers alone. Comparing the effects of deployments in the field will require task-specific benchmarks and measurement methods.

Worken focuses on treating the scale of robot ownership and shipments separately from the amount of work robots produce in the field. Manufacturing and deployment numbers indicate equipment supply and placement, but alone they do not show how long robots work autonomously or whether their output meets required standards. The framework’s combination of operating robot count, autonomous operating time and output efficiency ratio appears to be an attempt to include post-deployment operations in the basis for evaluation. However, Worken is not currently an established industry standard, and AGIBOT itself describes it as a work in progress. Because it refers to human work benchmarks for specific tasks, the comparability of the metric will depend on what constitutes acceptable quality and how differences in work environments and tasks are accounted for. How each element of the formula is measured, and how manual intervention or work interruptions are handled, are also issues relevant to evaluations in practice. A related article by this newspaper, “humanoid.guide reports on developments in AgiBot revenue and humanoid shipments”, covered developments in revenue and shipments. Revenue and shipments do not measure the same outcomes as Worken. The former are indicators of business scale and supply performance, while Worken is a proposal to measure autonomy and task output after deployment. Considering the two side by side still does not directly establish the amount of work performed in real-world settings. Business indicators and operating performance need to be evaluated separately. The company also noted that software and AI model improvements could enable the same robot to perform more work over its period of use, and outlined an approach that considers outcomes alongside life cycle costs. This suggests a direction in which value is not fixed at the point of deployment, but changes in performance during operation are also included in evaluations. However, the published explanation did not provide measured Worken values or a common conversion benchmark applicable across different tasks. Methods for setting task-specific benchmarks, verifying actual operating data and measuring performance while accounting for human intervention will help determine the metric’s usefulness.

AGIBOT describes Worken as “effective intelligent work hours.” The number of operating robots refers to robots performing tasks in the field, not manufactured or in-stock units. Autonomous operating time means the time spent working autonomously, rather than the time the robot is powered on. For the output efficiency ratio, the company evaluates the amount of work that meets acceptable quality standards, using benchmarks such as human performance for specific tasks. AGIBOT said it would improve the framework based on deployment and operating data and develop task-specific benchmarks. At the event, AGIBOT also unveiled AHA3, its third-generation hardware architecture for humanoid robots. It also showcased GO-3 Preview, based on WRAM (World Reasoning Action Model), which integrates inference, prediction and action, and introduced the industrial robot G2 Max. The company said the robot has 21 degrees of freedom, a maximum payload capacity of 40 kilograms across both arms, and an adjustable working height of 0.1 to 2.5 meters.