Industry observers might not have anticipated a brand-new framework for evaluating robotic productivity, yet the landscape is shifting. As the world's leading manufacturer of humanoid robots by shipment volume, Agibot is urging the global market to pivot away from merely tallying deployed units and toward quantifying actual labor. Unveiled during the company's annual Innovation Day showcase in Shanghai, the newly proposed metric dubbed “Worken”—a portmanteau of “work” and “token”—seeks to establish a universal standard for useful output delivered by operational machines.

Alongside this metric, the firm rolled out a robust portfolio of innovations. Highlights include the heavy-duty G2 Max semi-humanoid industrial system, featuring a wheeled base supporting a legless upper human body, alongside the advanced AHA3 hardware architecture, a sneak peek at the GO-3 foundation model, and an upcoming fully open-source robotic operating system scheduled for release before year's end.

The Growing Need for Labor Metrics in Robotics

Establishing clear benchmarks for automated labor has become an operational necessity. While Agibot currently paces global shipments according to recent market metrics, the company recognizes a fundamental complication: robotic output is hardly uniform. Operations range from fully autonomous execution to partial automation, remote teleoperation, and processes requiring constant human oversight.

“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?” noted Peng Zhihui, co-founder, president, and CTO of Agibot. “Worken is our attempt to develop a new way of answering that question. We want to move beyond measuring how many robots are built or delivered and begin measuring how effectively they can perform useful work, autonomously and consistently, in the real world.”

Calculating the “Effective Intelligent Work Hour”

The mathematics behind the proposal remain straightforward. Formally defined as an “effective intelligent work hour,” a Worken is computed via a specific equation:

  • Number of Active Robots
  • Multiplied by Autonomous Working Hours
  • Multiplied by Output Efficiency Ratio

Active robots denote units actively performing assigned duties, while autonomous working hours track duration free from ongoing human intervention. The output efficiency ratio evaluates completed acceptable output against predefined performance baselines, such as standard hourly production targets or average human benchmarks operating under identical conditions.

Put simply, a fleet of 10 machines operating autonomously for six hours at an 80 percent efficiency threshold yields a total of 48 Workens. This mirrors a full-time-equivalent (FTE) labor model familiar to corporate financial officers, while the incorporation of “token” nods to the metered, billed output models popularized by modern large language models.

Expanding the Hardware and Software Ecosystem

Beyond performance metrics, Agibot's Shanghai showcase featured extensive technological expansions. Taking center stage was the G2 Max, engineered specifically for demanding warehouse and manufacturing environments. Supported by a mobile platform, the machine boasts 21 degrees of freedom, a dual-arm payload capacity reaching 40 kilograms, and an adjustable operating height spanning from 0.1 to 2.5 meters to handle floor-level tasks to overhead retrieval.

Powering these systems is AHA3, the enterprise's third-generation humanoid framework. Integrating movement, communication, and manipulation into a cohesive platform termed “Three Intelligences in One,” the architecture incorporates integrated computing, Wi-Fi 6, 5G connectivity, joint liquid-cooling mechanisms to prevent thermal overload, and IP67-rated environmental sealing.

Advanced Grippers and Foundation Models

Agibot also upgraded its dexterous manipulation tools through the OmniHand 3 series—available in Lite, standard, and Ultra variants—alongside the OmniPicker 3 tactile gripper designed for non-articulated tasks. To capture human nuance for training data, the company introduced the MEgo Glove, a wearable sensor capturing hand and force subtleties during real-world tasks.

On the artificial intelligence front, the enterprise previewed its GO-3 foundation model, driven by a proprietary World Reasoning Action Model (WRAM). Additional software introductions featured the AGILE agentic self-learning system and WITA-Omni 1.0, a full-duplex conversational model operating locally on the hardware to allow natural, uninterrupted verbal interaction.