AGIBOT has introduced Worken, a proposed framework designed to measure the productivity of robots in real-world environments, during its Innovation Day 2026 event in Shanghai. Combining the concepts of "Work" and "Token," the initiative aims to shift industry focus from manufacturing volume to the amount of useful work robots can perform autonomously and consistently. Peng Zhihui, co-founder and CTO of AGIBOT, stated that while hardware and AI models have advanced significantly, the industry must now address how much productive work deployed robots actually deliver in factories, warehouses, and service settings.
The Worken framework defines productivity as "effective intelligent work hours," calculated by multiplying the number of active robots by their autonomous working hours and an output efficiency ratio. This metric assesses productive output against reference benchmarks, such as human worker standards, rather than simply counting units manufactured or powered-on time. AGIBOT described this approach as a way to evaluate the economic value of robotics, noting that software improvements can increase a machine's useful output over its lifecycle. The company plans to refine Worken through operational data and task-specific benchmarks to serve as a reference for physical AI value.
Alongside the new metric, AGIBOT unveiled several technological advancements. The company introduced AHA3, an integrated hardware platform unifying locomotion, interaction, and manipulation intelligence. Built on motor-drive technology and high-speed connectivity including Wi-Fi 6 and 5G, AHA3 features a joint liquid-cooling system and IP67-rated joints to support continuous industrial operation. In the realm of motion intelligence, AGIBOT highlighted updates to its AGILE framework, which now includes a self-evolving agent capability designed to help robots adapt to changing site conditions beyond predefined behaviors.
The event also featured the debut of GO-3 Preview, a next-generation embodied AI foundation model built around a World Reasoning Action Model (WRAM). Designed to anticipate action consequences and generalize across tasks, GO-3 Preview reportedly ranked first in task success rates on RoboDojo and achieved top scores on RoboTwin and RoboColiseum benchmarks. For human-robot interaction, the company showcased WITA-Omni 1.0, an omni-modal model integrating visual perception, speech, and physical actions into a continuous system that currently leads on embodied interaction benchmarks like Daily-Omni.
On the hardware front, AGIBOT launched the G2 Max, a heavy-duty industrial robot featuring 21 degrees of freedom and a dual-arm handling capacity of up to 40 kg. Designed for manufacturing and logistics, the robot supports force-controlled manipulation and omnidirectional mobility. The company demonstrated real-world deployments of its G2 family in tablet production, semiconductor packaging, CNC machine tending, and logistics operations at clients like Longcheer. Beyond industry, AGIBOT highlighted service applications in transportation and tourism, including deployments at Guangzhou Metro and the Chimelong complex.
To support these systems, AGIBOT released Link-U OS 1.0, described as the first operating system purpose-built for embodied intelligence. Built on a unified hardware abstraction layer and an embodied intelligence-native kernel, the OS connects robot hardware with upper-layer applications. The company announced plans to make Link-U OS fully open-source by the end of 2026 and has donated the project to the OpenAtom Foundation to foster a collaborative humanoid robot community. AGIBOT's ecosystem currently includes partnerships with 255 universities, 511 enterprises, and over 80,000 developers.
Sources: The Robot Report.