Google DeepMind Releases Gemini 2 Humanoid Model

Details

Google DeepMind has released a Gemini 2-based model targeting humanoid robot control, according to AI Business. Google frames the launch as progress toward what it calls 'physical AGI' — a term implying general-purpose robot intelligence capable of operating in unstructured real-world environments rather than constrained industrial settings.\n\nDetails in the available content are sparse: the report acknowledges that 'safety and real-world deployment remain significant hurdles,' suggesting the model is at an early or controlled-research stage rather than production-ready. No specific benchmarks, hardware platforms, or deployment partners are mentioned in the extracted content.\n\nThis builds on Google DeepMind's prior robotics research (RT-2, SayCan, and related foundation models for robots). A Gemini 2-class model applied to humanoid control would represent a meaningful step — Gemini 2 being a frontier multimodal model gives it stronger scene understanding and instruction-following than prior robotics-specific models. Independent technical details have not been confirmed at time of writing.