According to reports, Zhiyuan Robotics officially launched a new embodied foundation large model Genie Operator-2 (GO-2). The core mission of this model is to completely bridge the last mile from "logical reasoning" to "precise action execution", allowing robots not only to understand instructions but also to reliably complete tasks.

image.png

Core Black Technology: The First "Action Chain-of-Thought"

GO-2 introduces a brand-new interaction logic, completely changing the way robots think:

  • Rejecting "Watch and Do": Unlike traditional models that directly output control signals, GO-2 first generates a high-level action sequence in its mind as an overall task plan.

  • Think Clearly Before Acting: Through this structured intermediate representation, the robot achieves a leap from "perceptual input" to "action space reasoning". This technology has been accepted by the top AI conference CVPR 2026.

Architecture Innovation: Asynchronous Dual System Ensures "No Deviation in Execution"

To address stability issues during robot execution, Zhiyuan built a precise and coordinated asynchronous system:

  • Slow System (Brain): Runs at a low frequency, continuously outputs long-range logical action planning "intention flow".

  • Fast System (Muscle): Runs at a high frequency, real-time aligns with the slow system's plan. When encountering desk height deviations or object sliding, the fast system makes instant corrections to ensure the final action follows the preset trajectory.

Outstanding Achievements: Multiple Benchmark Tests Set SOTA Records

In various authoritative tests measuring embodied intelligence capabilities, GO-2 demonstrated dominant performance:

  • LIBERO Benchmark: Average success rate reached 98.5%, ranking first globally in four core tasks.

  • Genie Sim3.0: With training only on simulation data, the success rate in real environment testing reached 82.9%, significantly outperforming competing models (such as π0.5).

Industrial Vision: From "Laboratory" to "Production Line"

Zhiyuan Robotics does not just release a model, but builds a closed-loop learning system:

  • Distributed Evolution: Relying on the Genie Studio platform, the model continuously collects interaction data in real environments and optimizes online, achieving "smarter with use".

  • General Brain: GO-2 aims to become a general brain for embodied intelligence, helping robots smoothly transition from virtual simulations to complex industrial application scenarios.

Conclusion: Opening the Era of "Unity of Knowledge and Action" in Robotics

As Zhiyuan Official stated, the ultimate pursuit of embodied intelligence is the unification of understanding and execution. With the launch of GO-2, robots are evolving from "thinkers" who only talk about theory into "practitioners" capable of precisely changing the physical world.