DeepMind Launches Gemini Robotics 2.0: A Single VLA Model for Physical Dexterity

DeryaTR_ · x · 2026-07-30

Google DeepMind has introduced Gemini Robotics 2.0, a significant release aimed at advancing robotics capabilities.

The core breakthrough of this model is that it serves as a single Vision-Language-Action (VLA) model. It unlocks physical dexterity across different end effectors—whether hands or grippers—using just one model checkpoint.

In practical demonstrations, Apptronik's Apollo 2 humanoid robot, equipped with the 22-DoF SharpaWave hand, successfully performed highly delicate tasks such as tying knots and sealing a ziplock bag.

Related event: Google DeepMind Unveils Gemini Robotics 2(20 posts)→

Original post →

More from Embodied

Embodied channel →