Gemini Robotics 2 pairs whole-body control with a planning model

Google DeepMind introduced Gemini Robotics 2 on 30 July, including a vision-language-action model, an embodied-reasoning model and an on-device variant. The system is designed to connect visual and language input with robot movement, including whole-body control and coordinated tasks.

The companion ER 2 model handles higher-level planning, communication and tool calls before delegating execution to lower-level control systems. Developers gained public access through the Gemini API and AI Studio, with private enterprise preview. DeepMind’s demonstrations include multi-robot work and adaptation to different bodies; they are demonstrations of model research, not a general consumer robot launch. Separating planning from motor control also clarifies what an integration must supply: physical hardware, control interfaces and task-specific safety constraints.