Gemini Robotics 2 can now command a humanoid robot from feet to fingertips, enabling more complex, real‑world tasks.

Google DeepMind’s latest breakthrough, Gemini Robotics 2, can now command a humanoid robot’s entire body—from the feet up to the fingertips—paving the way for more intricate, real‑world applications.

Whole‑Body Control Achieved

The new model extends the capabilities of its predecessor by integrating locomotion, balance, and fine motor control into a single neural network, allowing the robot to coordinate walking, reaching, and manipulating objects simultaneously.

Researchers demonstrated the system on a Boston Dynamics‑style humanoid, where it successfully navigated uneven terrain, climbed stairs, and performed delicate tasks such as picking up a cup without dropping it.

Technical Foundations

Gemini Robotics 2 builds on DeepMind’s Gemini series, combining large‑scale language models with reinforcement‑learning‑based motor primitives. The model receives sensory input from cameras and proprioceptive sensors, processes it through a transformer architecture, and outputs joint torque commands in real time.

Unlike earlier approaches that required separate modules for navigation and manipulation, this unified framework learns a shared representation of the robot’s body and environment, reducing latency and improving coordination.

Potential Applications

The ability to control a robot’s whole body opens doors for deployment in settings that demand both mobility and dexterity, such as warehouse automation, disaster response, and assisted living.

  • Dynamic inventory sorting in cluttered storage facilities
  • Search‑and‑rescue missions navigating rubble while handling tools
  • Personal assistance tasks like fetching items in homes

While the technology is still in research labs, DeepMind plans to collaborate with robotics manufacturers to integrate Gemini Robotics 2 into commercial platforms over the next few years.

“This is a significant step toward robots that can move and interact with the world as fluidly as humans do,” said a DeepMind spokesperson.

The development also raises discussions about safety protocols and ethical considerations, prompting calls for transparent testing standards before widespread adoption.

For a detailed look at the research and demonstration videos, see The Verge coverage of Gemini Robotics 2.