.svg){kind=spacer}
Details
- Google DeepMind has announced Gemini Robotics and Gemini Robotics-ER, extending Gemini 2.0’s capabilities to include advanced vision-language-action control for robots.
- The initiative features partnerships with Apptronik for humanoid robotics and includes testing by industry leaders such as Boston Dynamics and Agile Robots.
- Gemini Robotics enables direct physical action from AI models, while Robotics-ER boosts spatial reasoning for complex tasks including 3D mapping and grasp detection.
- The new models demonstrate two to three times better performance in generalization and safety over previous iterations.
- DeepMind also released the ASIMOV dataset to support robotics safety research and introduced constitution-based alignment frameworks for safer robot behavior.
Impact
With the debut of these models, Google DeepMind strengthens its competitive position against major players in the humanoid robotics sector. The models' versatility and robust safety systems are poised to accelerate real-world robotics adoption, potentially setting new industry standards for safe and adaptive AI in physical environments.
