Can robots think for themselves from head to toe? The future brought by 'Gemini Robotics 2'

Imagine this: you wake up in the morning and walk into the living room to find toys scattered everywhere and a pile of laundry. In the past, you might have just sighed, but now you can simply say to the humanoid robot in the living room, “Can you tidy up the room?” The robot doesn’t just move its arms; it walks, crouches, and stretches, skillfully organizing the toys. If the cleaning seems like it might take too long, another nearby robot approaches and collaborates, finishing the job in an instant.

This is not a story from the distant future. It is a scene that the new physical AI system unveiled by Google DeepMind, ‘Gemini Robotics 2,’ aims to make a reality. [Source 1]

Why is this important?

Until now, most robots we have seen were either fixed in a specific location or moved only their upper bodies to perform simple, pre-programmed tasks. [Source 6] However, human movement is much more precise and complex. When we pick something up, we don’t just use our arms; we balance with our feet, bend at the waist, adjust our fingers minutely, and use our whole body organically.

The reason Gemini Robotics 2 is attracting attention is that it provides robots with an intelligence similar to this ‘human movement.’ This allows robots to make their own judgments and move even in complex environments. Google DeepMind evaluates this system as a step toward the era of ‘Physical AGI,’ where robots understand and act within the physical world. [Source 11, Source 14]

Easy understanding: What is ‘whole-body intelligence’?

To use a simple analogy, if existing robots were ‘automated dolls that only repeat set movements,’ robots equipped with Gemini Robotics 2 are like ‘professional athletes who perfectly understand and control their bodies.’

Roboticists call this ‘whole-body control’ or ‘loco-manipulation’ (performing movement and manipulation simultaneously). [Source 8] Gemini Robotics 2 possesses a core brain model called ‘Gemini Robotics ER 2,’ which performs ‘embodied reasoning.’ [Source 4, Source 7]

The robot sees its surroundings with cameras, assesses the situation through the AI model, and then sends commands to every joint from its toes to its fingertips to perform complex actions. [Source 7, Source 10] Furthermore, multiple robots can communicate and collaborate based on shared knowledge, maximizing work efficiency. [Source 15]

Current status: How advanced is it?

Gemini Robotics 2 has currently been released in three different models depending on performance and purpose. [Source 10] Compared to previous models, this system has achieved a significant technological leap, capable of understanding much more complex environments and performing multi-step tasks. [Source 4, Source 7]

It is already proving its capabilities in real-world scenarios. For example, robots of different types have been demonstrated communicating and working together via Gemini Robotics ER 2. [Source 15] However, experts point out that verifying safety in highly uncertain and unpredictable environments, such as actual homes, is a challenge we must solve going forward. [Source 11]

The future with robots

We are rapidly moving into an era where robots grasp situations on their own, interact naturally with humans, and solve complex problems. Gemini Robotics 2 is expected to be the core technology of this transition. [Source 13] What we should pay attention to going forward is how safely and familiarly this technology is introduced into our daily lives. The day is not far off when a robot will become a smart, reliable partner that helps with your daily life.

References

1. Gemini Robotics 2 brings whole body intelligence to robots
2. Gemini Robotics 2 brings whole body intelligence to… - YouTube
3. Google News - Google DeepMind releases Gemini Robotics 2 system…
4. Google reveals Gemini Robotics 2.0, promising… - Ars Technica
5. Google DeepMind Releases Gemini Robotics ER 2 · Digg
6. Google DeepMind’s Gemini Robotics 2 controls whole humanoids
7. Google DeepMind’s new AI model can control a robot’s entire body
8. Gemini Robotics 2 Controls Full Humanoids: Legs, Torso, Arms …
9. Google DeepMind Ships Three Physical AI Models For Whole Body …
10. Gemini Robotics 2 Brings Google’s AI Into the Physical World …
11. Techmeme: Google DeepMind releases Gemini Robotics 2, which…
12. Gemini Robotics: Google’s AI Breakthrough That’s Bringing Robots…
13. Gemini Robotics: Google’s AI Push Into Humanoid Robots
14. Gemini Robotics ER 2