Google DeepMind’s new AI model can control a robot’s entire body

Summary

Google DeepMind's Gemini Robotics 2 model now enables humanoid robots to perform the full range of bodily movements, from walking and squatting to fine finger motions.

Why it matters

This advancement expands the functional repertoire of robots, allowing them to tackle complex real-world tasks that require coordinated movement across the entire body. The integration of embodied reasoning and on-device processing also enhances autonomy, safety, and adaptability across different robot designs.

Highlights

• Control of whole-body movements such as walking and squatting.
• Increased dexterity for complex tasks like unscrewing a light bulb with five-fingered hands.
• Improved human perception and task-boundary awareness, boosting safety and multi-robot collaboration.

Critical take

While Gemini Robotics 2 is a significant step forward, the article notes that its movement speed still lags behind human capabilities and that practical deployment may be limited in dynamic environments.