leaderobot.com Publishes Opinion Article on Computing Resources for Embodied AI

leaderobot.com published an opinion article on computing resources for embodied AI on Nov. 3, 2026.

With the definition and measurement methods for “Worken” reported to be unsettled, comparing the amount of work robots perform requires checking not only the metric’s formula but also the conditions under which it is measured. Opinion articles about computing resources also need to be evaluated separately from official announcements and operational data.

This article is an opinion piece addressing issues surrounding embodied AI; it is not an official company announcement or a published performance assessment. Its publication therefore cannot be considered independent substantiation of the computing resources required or robots’ real-world operational performance. Readers should consult the original sources to assess individual arguments and figures. Related articles in this newspaper have outlined the formula for the “Worken” operational-volume metric reportedly proposed by Zhihui Peng of AgiBot — number of robots in operation, autonomous working hours and output efficiency ratio — and noted that its definition and measurement methods remain unsettled AgiBot reportedly proposes “Worken,” a metric for robot operations, leaderobot.com reports. An article on NVIDIA Cosmos also identifies adapting models to specific sensors, tasks and data as a development challenge NVIDIA explains world models for interactive simulation using Cosmos. These concern different issues — how to measure operational volume and how to adapt models during development — and cannot be combined into a single metric. For this reason, discussions of computing resources need to distinguish between the computing used for model training and simulation and the results of robots’ work in real-world environments. The scale of computing alone does not show what tasks robots can perform autonomously or under what measurement conditions. Conversely, operational metrics alone cannot determine the scale of computing infrastructure needed to train or adapt models. Connecting these assessments requires published information specifying the tasks, measurement methods and operating conditions in real-world environments. Individual figures and claims in opinion articles should also be checked against the source and calculation conditions, and treated separately from company announcements and demonstration data. The next points to verify are the metric’s definition and measurement conditions, results from real-world operations, and how those results relate to computing requirements.