Google’s new Gemini AI models could teach robots to think and act like us

Google’s Gemini has been the frontrunner along with OpenAI’s ChatGPT and Microsoft’s Copilot for a while now. From integrating AI assistance in smartphones to weaving it into its services, Gemini has drastically evolved since its conception in 2023.

However, this time around, Google just dropped two game-changing AI models that could potentially revolutionise how robots interact with our world. The tech giant’s DeepMind team unveiled Gemini Robotics and Gemini Robotics-ER which are designed to give robots human-like abilities to see, think, and act. Here’s everything we know so far.

What is Google's Gemini Robotics, and what can it do?

According to the brand, Gemini Robotics is an advanced vision-language-action (VLA) model built on Gemini 2.0. It explains how physical actions are added in this model as a new output capability for directly controlling robots.

The Gemini Robotics model is also said to have major improvements in three crucial areas: generality (adapting to new situations), interactivity (engaging with people and environments), and dexterity (performing delicate tasks like folding paper or unscrewing bottle caps).

ALSO READ: Google’s Gemini 2.0 is now available to everyone

Google also mentions how while it is trained primarily on the bi-arm ALOHA 2 platform, it will work seamlessly with other systems, like the Franka arms, which are currently employed across several research labs.

What is Google's Gemini Robotics-ER, and what can it do?

Gemini Robotics-ER, on the other hand, takes things up a notch by enhancing Gemini’s understanding of the world. The brand mentions how this particularly includes spatial reasoning.

Google states how this model significantly improves certain abilities; these include pointing as well as 3D detection. By combining spatial reasoning with Gemini’s coding capabilities, it can also quickly develop new functions on the fly. For instance, when shown a coffee mug, it can instantly determine how to grasp it with two fingers on the handle and plan a safe approach path.

ALSO READ: How to chat with Gemini from your iPhone’s lock screen

What’s more, the Gemini Robotics-ER handles all necessary steps required to control a robot immediately, like perception, state estimation, spatial understanding, planning, and even code generation. Although a thing of the near future, Google goes on to state how both models are specifically designed for factory systems, warehouse automation, and humanoid robots.

What does this mean for the future?

Google’s developments certainly blur the line between sci-fi and reality, raising fascinating questions (and concerns!) about how robots might transform industries in the future. But if that’s not all, they will also potentially impact our daily lives with their increasingly human-like abilities.

With robots on the route to understanding and interacting with us more naturally, are we finally witnessing the birth of truly intelligent machines? Perhaps – but only time will tell!