AgiBot outlines 2026 embodied intelligence industry report, explains XYZ curve and Worken

AgiBot Innovation (Shanghai) Technology Co., Ltd. published an article on November 9, 2026, outlining key points from the 2026 Embodied Intelligence Industry Innovation Report. The report was released on October 29, 2026, by Xinhua News Agency’s China Economic Information Service and the China Academy of Information and Communications Technology’s Shanghai Industrial Innovation Center. It organizes industry stages using the “XYZ curve,” and sets out “three intelligences in one” as a technology approach and “effective intelligent work hours (Worken)” as a productivity metric.

The report distinguishes manufacturing and shipment volumes from actual productivity, proposing that robots that work continuously in real-world environments be evaluated using Worken. For businesses considering deployments, the focus of evaluation is not only the number of units installed, but also how to measure autonomous working time and output efficiency.

The report’s central idea is to shift evaluation from the number of robots manufactured or shipped to their ability to work continuously and deliver results in the field. Its approach to Worken, based on the number of active robots, autonomous working time and output-efficiency ratio, aims to assess real-world work that deployment numbers alone cannot capture. However, identifying the components of a metric is a separate challenge from establishing a measurement standard that enables comparisons across different worksites and tasks. AgiBot’s article mentions a Worken original sample database, but does not provide sample figures or methods for calculating and validating each component. For now, the proposal should be read as a common metric; the comparability of actual results needs to be assessed separately. The XYZ curve places the X phase in 2022-2025, the Y phase in 2026-2030 and the Z phase after 2030. This is a framework for viewing the industry stages set out in the report and should be distinguished from objective criteria for determining when the industry moves between stages or forecasts of adoption levels during those periods. By presenting Worken, which measures actual deployment, alongside the XYZ curve, which represents stages of development, the report offers a framework for describing industry progress in terms of both “which stage” and “what has been produced in the field.” The “three intelligences in one” concept puts the emphasis on coordinating motion, task and interaction intelligence, in addition to the robot body itself. The report also treats data feedback loops, operations and commercial ecosystems as part of deployment capability, suggesting that implementation should be assessed at the system level, including the body, models, data and operations, rather than at the level of the robot alone. Physical AI News covered AgiBot, Tsinghua University and others unveil CEL, testing collective learning with 96 robots, which addressed a system for robots to share experience. This is related to the report’s discussion of data circulation, but does not mean that the report’s framework was validated in that experiment. Physical AI News has also covered AgiBot proposes Worken, a metric for measuring robot activity, leaderobot.com reports; the latest article incorporates the company’s approach to evaluating actual work into its discussion of industry stages. The industry implications lie in a phased approach: start with transport, sorting and patrol inspections, then apply operational experience in the field to new uses. Before expanding applications, companies will need to determine how to record continuous operation, task quality and operating burden in each setting, and how to reflect them in Worken’s “output-efficiency ratio.” Another point for testing is the conditions under which coordination among different types of intelligence and data circulation work as robots move into more complex environments and tasks. Whether the report’s framework becomes a common industry benchmark will depend on how clearly the metric and measurement procedures are defined and how much deployment data is made available.

The report divides the period from 2022-2025 into the X curve, or “development and trial,” 2026-2030 into the Y curve, or “deployment and growth,” and the period after 2030 into the Z curve, or “deployment and adoption.” Its “three intelligences in one” technology concept links the robot body with motion intelligence, task intelligence and interaction intelligence. Worken is modeled using three components: the number of actively operating robots, autonomous working time and output-efficiency ratio. For field deployment, the report identifies high-confidence industrial applications such as transport, sorting and patrol inspections as short-term targets; commercial services and complex industrial applications as medium-term targets; and all scenarios as long-term targets. AgiBot’s article describes a phased approach that applies data and experience accumulated in earlier uses to subsequent ones, gradually broadening the scope of applications.