Zhiyuan Unveils GO-2 Embodied Model with Action Thought Chain, Setting New Industry SOTA
Zhiyuan Robotics has introduced a new embodied foundation model called Genie Operator-2 (GO-2). Its core mission is to close the gap between logical reasoning and precise action execution, enabling robots to understand commands and reliably carry out tasks.

Key Innovation: The First Action Chain-of-Thought
GO-2 introduces a fresh interaction logic that fundamentally changes how robots think:
Rejecting "Watch and Do": Unlike traditional models that output control signals directly, GO-2 first mentally generates a high-level action sequence as a complete task plan.
Think Clearly Before Acting: Using this structured intermediate representation, robots leap from perception to action space reasoning. This technology has been accepted by the top AI conference CVPR 2026.
Architecture Innovation: Asynchronous Dual System Ensures Precision
To address stability issues during execution, Zhiyuan built a precise asynchronous system with two coordinated components:
Slow System (Brain): Operates at a low frequency, continuously producing long-range logical action plans or "intention flow."
Fast System (Muscle): Runs at a high frequency, aligning in real time with the slow system. When desk height changes or objects slip, the fast system instantly corrects to keep the action on the intended path.
Outstanding Results: SOTA Records Across Multiple Benchmarks
In authoritative tests measuring embodied intelligence, GO-2 showed dominant performance:
LIBERO Benchmark: Achieved an average success rate of 98.5%, ranking first globally in four core tasks.
Genie Sim3.0: Trained only on simulation data, it reached an 82.9% success rate in real-world testing, significantly outperforming competitors like π0.5.
Industrial Vision: From Lab to Production Line
Zhiyuan Robotics is building more than a model — it's creating a closed-loop learning system:
Distributed Evolution: Through the Genie Studio platform, the model continuously collects interaction data in real environments and optimizes online, becoming smarter with use.
General Brain: GO-2 aims to serve as a general brain for embodied intelligence, helping robots transition smoothly from virtual simulations to complex industrial settings.
Conclusion: Ushering in the Era of Unity of Knowledge and Action
As Zhiyuan Official stated, the ultimate goal of embodied intelligence is to unify understanding and execution. With the launch of GO-2, robots are evolving from thinkers who only talk theory into doers that can precisely change the physical world.
Related article
Stripe to Acquire OpenRouter as AI Model Routing Gains Momentum
Stripe has agreed to acquire OpenRouter, an AI model-routing platform that gives developers access to hundreds of models through a single interface. The deal adds model selection and routing to Stripe’s existing work around AI usage and token-based b
Former OpenAI Researcher Barrett Zoph Returns to Google to Support Gemini Development
Barret Zoph, who previously abruptly left and joined OpenAI due to a disagreement with Thinking Machines Lab CEO Miriam Mularati, has once again changed employers. Alphabet's Google announced that Zoph will return as Vice President of Research, bring
How Recursive Intelligence Raised $335M at $4B Valuation in 4 Months
The co-founders of startup Ricursive Intelligence seemed destined to be co-founders.Anna Goldie, CEO, and Azalia Mirhoseini, CTO, are so well-known in the AI community that they were among those AI engineers who “got those weird emails from Zuckerber
Related Special Topic Recommendations
Comments (0)
0/500
Zhiyuan Robotics has introduced a new embodied foundation model called Genie Operator-2 (GO-2). Its core mission is to close the gap between logical reasoning and precise action execution, enabling robots to understand commands and reliably carry out tasks.

Key Innovation: The First Action Chain-of-Thought
Rejecting "Watch and Do": Unlike traditional models that output control signals directly, GO-2 first mentally generates a high-level action sequence as a complete task plan.
Think Clearly Before Acting: Using this structured intermediate representation, robots leap from perception to action space reasoning. This technology has been accepted by the top AI conference CVPR 2026.
Architecture Innovation: Asynchronous Dual System Ensures Precision
To address stability issues during execution,
Slow System (Brain): Operates at a low frequency, continuously producing long-range logical action plans or "intention flow."
Fast System (Muscle): Runs at a high frequency, aligning in real time with the slow system. When desk height changes or objects slip, the fast system instantly corrects to keep the action on the intended path.
Outstanding Results: SOTA Records Across Multiple Benchmarks
In authoritative tests measuring embodied intelligence,
LIBERO Benchmark: Achieved an average success rate of 98.5%, ranking first globally in four core tasks.
Genie Sim3.0: Trained only on simulation data, it reached an 82.9% success rate in real-world testing, significantly outperforming competitors like π0.5.
Industrial Vision: From Lab to Production Line
Distributed Evolution: Through the
General Brain:
Conclusion: Ushering in the Era of Unity of Knowledge and Action
As
Stripe to Acquire OpenRouter as AI Model Routing Gains Momentum
Stripe has agreed to acquire OpenRouter, an AI model-routing platform that gives developers access to hundreds of models through a single interface. The deal adds model selection and routing to Stripe’s existing work around AI usage and token-based b
Former OpenAI Researcher Barrett Zoph Returns to Google to Support Gemini Development
Barret Zoph, who previously abruptly left and joined OpenAI due to a disagreement with Thinking Machines Lab CEO Miriam Mularati, has once again changed employers. Alphabet's Google announced that Zoph will return as Vice President of Research, bring
How Recursive Intelligence Raised $335M at $4B Valuation in 4 Months
The co-founders of startup Ricursive Intelligence seemed destined to be co-founders.Anna Goldie, CEO, and Azalia Mirhoseini, CTO, are so well-known in the AI community that they were among those AI engineers who “got those weird emails from Zuckerber





Home






