Zhiyuan Robotics Unveils GO-2: A Leap Forward in Embodied AI for Robots

Daily Technology

10/04/2026

Zhiyuan Robotics has officially launched its next-generation embodied foundation model, Genie Operator-2 (GO-2). This advanced AI aims to bridge the critical gap between a robot's ability to reason logically and its capacity for precise physical action, promising more stable and reliable robotic performance in real-world scenarios.

Key Takeaways

GO-2 unifies logical reasoning and precise action execution within a single architecture.
The model surpasses state-of-the-art performance on multiple robotic benchmarks.
It addresses the "last mile" problem, ensuring robot actions consistently follow planned tasks.
GO-2 represents a shift from robots merely understanding to reliably acting in the physical world.

Bridging the Gap Between Thought and Action

Last year, Zhiyuan Robotics introduced the Genie Operator-1 (GO-1), which successfully integrated vision, language, and action, enabling robots to understand instructions, identify scenes, and plan tasks. However, a significant challenge emerged when deploying GO-1 in complex environments: while robots could generate reasonable plans, their physical execution often faltered.

Traditional embodied AI models often suffer from a fragmented process. High-level reasoning generates abstract instructions, which are then passed to a control system for execution. This creates a substantial disconnect between abstract commands and the precise movements required in the physical world. During execution, control modules can bypass planning, relying on immediate visual input, leading to accumulated errors and deviations from the intended plan, especially in long-horizon tasks.

The GO-2 Solution

GO-2 is engineered to solve this "last mile" problem. Its core objective is to empower robots not only to comprehend their surroundings but also to interact with them with unwavering stability and reliability. This marks a significant evolution from GO-1's "understanding" capabilities to GO-2's focus on "acting."

The model aims to transition from occasional successes to consistent, stable task completion. It moves beyond simply performing movements to enabling precise interaction and dependable deployment in physical environments. GO-2 is designed to serve as a general-purpose "brain" for embodied intelligence, fostering trustworthiness and deployability.

By breaking down the divide between semantics and action, GO-2 unifies planning and execution. This ensures that every robot movement is dynamically adapted to the complexities of the physical world, leading to stable operation in diverse real-world scenarios.

2026-04-01

Google Implements Gmail Username Change Functionality

Google now allows users to change their Gmail address without creating a new account. Learn about the process, its benefits, and key limitations for services.