Google DeepMind is launching two new artificial intelligence models designed to help robots "perform a wider range of real-world tasks than ever before." The first model, named Gemini Robotics, is a vision-language-action model capable of understanding new situations even without prior training.

Gemini Robotics is built on Gemini 2.0, the latest version of Google's flagship artificial intelligence model. During a press conference, Carolina Parada, Senior Director and Head of Robotics at Google DeepMind, stated that Gemini Robotics "leverages Gemini's understanding of the multimodal world and transfers it to the real world by adding physical actions as a new modality."

The new model has made advancements in three key areas that Google DeepMind considers crucial for creating useful robots: generalization, interactivity, and dexterity. In addition to its ability to generalize new scenarios, Gemini Robotics can better interact with people and the environment. It can also perform more precise physical tasks, such as folding a piece of paper or removing a bottle cap.

Google DeepMind's new Gemini Robotics model makes robots more dexterous.

Parada said, "In the past, we made progress in each of these areas individually, but now we are significantly improving performance across all three areas with a single model. This allows us to create robots that are more capable, responsive, and robust to environmental changes."

Google DeepMind also introduced Gemini Robotics-ER (Embodied Reasoning), which the company describes as an advanced vision-language model capable of "understanding our complex and ever-changing world.

As Parada explained, when you're packing a lunchbox with various items on the table in front of you, you need to know where everything is, how to open the lunchbox, how to pick up items, and where to place them. This is exactly the kind of reasoning Gemini Robotics-ER is designed to handle. It aims to enable robotics experts to connect with existing low-level controllers (systems that control robot movements), allowing them to enable new features powered by Gemini Robotics-ER.

Gemini Robotics can also assist robots in performing a range of tasks.

Regarding safety, Vikas Sindhwani, a researcher at Google DeepMind, told reporters that the company is developing a "layered approach" and added that the Gemini Robotics-ER model is "trained to assess whether executing potential actions in specific scenarios is safe." The company has also released new benchmarks and frameworks to help the AI industry further advance safety research. Last year, Google DeepMind introduced the "Robot Constitution," a set of rules for robots inspired by Isaac Asimov.

Google DeepMind is collaborating with Apptronik to "build the next generation of humanoid robots." Additionally, Google DeepMind is granting "trusted testers" access to its Gemini Robotics-ER model, including Agile Robots, Agility Robotics, Boston Dynamics, and Enchanted Tools. Parada said, "We are highly focused on building intelligence that can understand the physical world and take action within it. We are very excited to leverage this across multiple embodiments and applications."