AGIBOT has made a significant announcement at CES 2026 by launching Genie Sim 3.0, an innovative robot simulation platform aimed at enhancing the development of embodied intelligence in robotics. This release is not just another upgrade; it represents a comprehensive, open-source framework that facilitates the training and evaluation of intelligent robots, promising to revolutionize how we approach robotic learning and performance assessment.
One of the standout features of Genie Sim 3.0 is its seamless integration with NVIDIA Isaac Sim, leveraging the powerful capabilities of NVIDIA Omniverse. This partnership allows for a streamlined process that encompasses everything from digital asset creation and scene generation to data collection, automated evaluation, and physics-based simulation—all within a single pipeline. This means developers can now create and test their robotic models more efficiently than ever, significantly reducing their reliance on physical prototypes while accelerating the iteration of models.
At the heart of this launch lies the Genie Sim Benchmark, a robust evaluation system crafted specifically for assessing embodied intelligence. This benchmark covers an impressive array of over 200 tasks across more than 100,000 simulated scenarios. Its design aims to transcend simplistic metrics, allowing developers to construct comprehensive capability profiles for their robotic models, thereby providing deeper insights into performance.
In terms of data, AGIBOT asserts that Genie Sim 3.0 hosts the largest open-source simulation dataset available in the realm of embodied AI, boasting over 10,000 hours of synthetic data. Unlike traditional datasets that often rely on abstract test cases, this dataset mirrors real-world operational scenarios encountered by robots, making it invaluable for realistic training and evaluation. It incorporates various sensor modalities, including RGB-D inputs, stereo vision, and whole-body kinematics, and even features an automated data-collection toolkit that enhances the efficiency of gathering relevant information.
This platform also excels in low-latency teleoperation and scripted task execution, ensuring that robotics teams can operate smoothly and effectively. The inclusion of auto-annotation tools enables the system to label data automatically during collection processes, further streamlining workflow. Additionally, a built-in recovery mechanism allows tasks to be restarted after any failures, which AGIBOT claims significantly reduces both time and financial costs associated with dataset creation.
Another intriguing aspect of Genie Sim 3.0 is its language-driven scene creation feature. Users can describe environments using everyday language, prompting the system to generate structured scenes, visual previews, and a wide variety of semantic variations. This is powered by large language models that eliminate the need for manual logic scripting, making the process much more intuitive. Vision-language models refine these scenes to meet specific requirements, helping robots to adapt and generalize across diverse environments.
Moreover, the platform integrates 3D reconstruction with visual generation, utilizing handheld 3D laser scanners to capture real environments. These scanners combine RGB imagery, 360-degree LiDAR, and precise centimeter-level positioning. Remarkably, a single 60-second orbital video taken of an object can be transformed into a simulation-ready asset, drastically speeding up the process of scene construction.
Genie Sim 3.0 also has significant implications for industrial robotics, as AGIBOT positions it as the first platform to incorporate real industrial-scene datasets into both training and evaluation processes. This allows for the creation of one-to-one digital twins of logistics hubs, inspection sites, and production lines. Teams can rigorously test their models end-to-end without needing to deploy physical robots on-site, which not only accelerates validation cycles but also minimizes dependence on hardware.
All assets, datasets, and evaluation code within this simulation environment are fully open source, aligning with AGIBOT's strategy to establish a credible benchmark ecosystem. The company anticipates that researchers and industrial teams will leverage this platform to advance future standards in embodied AI.
As we delve into CES 2026, there’s an exciting array of developments to explore. For those eager to keep up with all the stories and coverage from the IE team, you can find more by following the links provided.
With a wealth of experience in technology journalism, Aamir has previously contributed to notable publications such as Exhibit Magazine, Republic World, and PR Newswire. His passion for technology and science drives his dedication to unpacking the latest innovations and exploring their impact on industries, lifestyles, and the future of humanity.
