Google DeepMind Launches Gemini Robotics ER 2, Positioned as the Strongest Embodied Reasoning Model
Google DeepMind has launched Gemini Robotics ER 2, positioned as the strongest embodied reasoning model currently available.
This model acts as the high-level brain for robots, understanding the real physical world, executing complex multi-step task planning and orchestration, progress tracking, and failure detection.
It can call tools such as Google Search and custom functions, directing underlying vision-language-action models to perform specific actions, while also supporting multi-robot collaboration.
Source: Public Information
ABAB AI Insight
Google DeepMind previously launched the Gemini Robotics series, including the VLA control model and earlier versions of ER. ER 2 is based on Gemini 3.5 Flash, enhancing spatial, temporal, and physical reasoning, and is open to AI Studio.
Capital and resources continue to tilt towards embodied intelligence, with DeepMind decoupling high-level planning from low-level execution, making it easier to adapt to different hardware forms like Apptronik and Boston Dynamics, reducing reliance on single machines.
Similar paths can be seen in Tesla's Optimus and Figure AI's end-to-end attempts, as the current robotics industry shifts from single-point action control to multi-machine collaboration and long-term task orchestration.
This essentially represents a restructuring of the industrial chain; after standardizing high-level reasoning models, hardware manufacturers focus on the execution layer, while the software layer controls task orchestration and tool invocation, accelerating AI penetration into the physical world.
ABAB News · Cognitive Laws
Brain planning determines the value of limbs
Decoupling execution allows for rapid body switching
Multi-machine collaboration is the breakthrough point for single-machine limits