Can fold paper and also slam dunk! Google releases a large model for robotic bases, significantly enhancing the versatility of robots

Wallstreetcn
2025.03.13 03:39
portai
I'm PortAI, I can summarize articles.

On March 12th, DeepMind, a subsidiary of Google, released a brand new AI model that specifically trained a model for controlling robots using its Gemini 2.0 language model. The adaptability to the environment, understanding of instructions, and operational flexibility have all been significantly improved

Another "milestone" in the development history of general-purpose robots has arrived.

On March 12, DeepMind, a subsidiary of Google, released a brand new AI model called Gemini Robotics and Gemini Robotics-ER. This model claims to enable robots to better adapt to complex environments and complete various high-difficulty tasks.

As of yesterday, the stock price of Google's parent company Alphabet Class C closed slightly up by 1.82%, at $169.00.

Can robots fold paper, tidy up desks, and dunk? Google DeepMind showcases impressive capabilities!

The AI model released by DeepMind has been further upgraded, and robots trained with the new model can not only smoothly fold paper foxes.

They can also tidy up desks according to voice commands.

They can even skillfully unzip zippers.

Finally, they can even perform a mini basketball dunk!

DeepMind has also partnered with the startup Apptronik to plan the creation of humanoid robots using this technology.

Netizens on X have expressed that having a robot butler in the future is really not a dream anymore, imagining asking a robot to make a cup of coffee.

In the past, the movements of robots required scientists to manually program them, which was time-consuming and labor-intensive. Now, thanks to new AI technology, robots can better adapt to their environments, and the speed at which they learn new skills has greatly accelerated. Kanishka Rao, Chief Software Engineer at Google DeepMind, stated:

The versatility of Gemini Robotics is twice that of the previous best model, achieving a significant leap towards general-purpose robots.

Gemini Robotics: Making Robots Smarter and More Flexible

It is reported that Google has specifically trained a model for controlling robots using its Gemini 2.0 language model. This endows robots with three major "superpowers":

  • Environmental Adaptability: Can quickly adapt to various new situations.
  • Instruction Comprehension: Can rapidly respond to verbal commands or changes in the environment.
  • Operational Flexibility: Possesses sufficient dexterity to manipulate objects.

This capability is crucial for the development of robotics technology, as robots previously performed excellently in laboratories but often struggled in uncontrolled environments.

A netizen commented that Gemini makes robots cooler and more interesting.

Ken Goldberg, a robotics professor at the University of California, Berkeley, stated:

This technology is undoubtedly an exciting advancement in the field of robotics, seemingly built on Google's advantages in large-scale data and computation.

He also mentioned that the most unique aspect of these new robotic models is their ability to run smoothly in the cloud, likely due to their access to Google's ultra-large language models that require substantial computational power. Russ Tedrake, a professor at MIT and vice president of robotics research at the Toyota Research Institute, believes:

Gemini is an important step.

However, he also stated:

There is still much work to be done for general-purpose robots to be ready for deployment.

Tech Giants Compete for Layout, Trillion-Dollar Market on the Verge of Takeoff

This move by DeepMind also signals that an "arms race" in the field of robotics is accelerating.

Bank of America analyst Ming Hsun Lee's team boldly predicts that humanoid robots are about to leave the proof-of-concept stage and will experience an explosion of applications across multiple industries within the next decade. It is expected that by 2030, global annual sales of humanoid robots will reach 1 million units, and by 2060, the total stock of humanoid robots will reach 3 billion units, equivalent to 0.3 robots per person at that time.

Tech giants including Tesla and OpenAI, along with numerous startups, are racing to develop AI "brains" in hopes of granting robots the ability to operate autonomously.

NVIDIA's CEO Jensen Huang even stated that this technology will create a massive market worth trillions of dollars, potentially "opening up the largest tech industry in history."