Gemini Robotics 2

Gemini Robotics 2, launched by Google DeepMind in July 2026, is an advanced physical AI model suite designed for general-purpose robotics. It bridges the gap between high-level reasoning and physical action using a Vision-Language-Action (VLA) model and an Embodied Reasoning (ER 2) model. It enables robots, including dual-arm and full humanoid platforms, to execute dexterous manipulation and whole-body coordination. Featuring advanced spatial intelligence, long-horizon planning, and real-time multi-robot collaboration, the models can be run locally on-device or streamed via the Gemini Live API for low-latency physical orchestration.

• End-to-end Vision-Language-Action (VLA) motor control [1.2.1]
• Whole-body coordination and humanoid/bi-arm dexterous manipulation
• Multi-robot collaboration and division of labor via ER 2
• Advanced spatial reasoning and long-horizon task planning
• On-device local deployment with real-time streaming API