DeepMind’s latest AI model can help robots fold origami and close Ziploc bags
Since its debut late last year, Gemini 2.0 continues to power a handful of Google products, including the new AI Mode Chatbot. Now, Google DeepMind is using the same technology for total fun. On Wednesday, AI Labs announced two new Gemini-based models that will “lay the foundation for a new generation of useful robots.”
The first is Gemini Robot Technology, designed by DeepMind to promote direct control of the robot. The company believes that robots’ AI systems need to be excellent in three qualities: generality, interactivity, and agility.
The first involves the flexibility of the robot to adapt to new situations, including those not covered by training. At the same time, interactivity encapsulates the robot’s ability to respond to people and the environment. Finally, there is flexibility, which is mostly self-evident: many people can complete many tasks without a second idea, involving fine motor skills that robots are difficult to master.
“Although our previous work has shown progress in these areas, Gemini robotics represents an important step in performance on all three axes, bringing us closer to a true universal robot,” DeepMind said.
For example, using a Gemini robot can power it, DeepMind’s Aloha 2 robot is able to fold origami and close Ziploc bags. The two-arm robot also understands all instructions in natural daily language. As can be seen from the video shared by Google, it can even complete tasks despite obstacles, for example, when the researcher moved around Tupperware, he just asked the robot to put the fruit in it.
Google is working with Apptronik, the company behind Apollo Double Pidel robots, to build the next generation of humanoid robots. Meanwhile, DeepMind is releasing Gemini robotics (or reflective reasoning). In the second model, the company said it will enable robotists to run their own programs using Gemini’s advanced reasoning abilities. DeepMind provides “trusted testers” including a Google subsidiary Boston Dynamics, access to the system.
