Google DeepMind demonstrated the Gemini Robotics 2 model controlling a humanoid robot and a dual-arm system to perform household chores this month [1].
This development represents a shift toward general-purpose robotics by integrating large-scale AI models with physical hardware. The ability to translate complex linguistic instructions into precise physical movements could accelerate the deployment of autonomous assistants in residential and commercial settings.
During the demonstrations, the AI controlled an Apptronik Apollo 2 humanoid robot and a separate dual-arm system [1]. The systems were shown performing tasks such as putting items away, showcasing the model's capacity for whole-body control and dexterity [2]. DeepMind said the platform includes safety guardrails to ensure robots operate securely within human environments [3].
One of the primary technical advantages of the system is its adaptability. Reports said the model can be adapted to a new robot with only a few hours of training [4]. This flexibility reduces the need for the exhaustive, robot-specific programming that has historically slowed the adoption of humanoid technology.
This release follows earlier strategic moves by the company to expand its robotics ecosystem. Google DeepMind previously entered a partnership with Agile Robots on March 24, 2026 [5]. That collaboration laid the groundwork for the integration of advanced AI models into mobile, humanoid platforms.
The Gemini Robotics 2 series focuses on bridging the gap between high-level reasoning and low-level motor control [2]. By utilizing a unified model, the robots can better understand the spatial relationships of objects, and the sequence of actions required to complete a multi-step chore [3].
“The model can be adapted to a new robot with only a few hours of training.”
The transition from task-specific programming to a generalized model like Gemini Robotics 2 suggests that humanoid robots may soon move from controlled lab environments to versatile real-world applications. By reducing the training time required for new hardware, Google is positioning its AI as a universal operating system for robotics, potentially standardizing how different robot brands interpret and execute physical commands.
