Google's new AI controls humanoid robots from feet to fingertips
What's the story
Google has unveiled the latest iteration of its Gemini Robotics artificial intelligence (AI) model, which can now control an entire humanoid robot. The previous version was limited to controlling only the upper body of a humanoid robot. The new and improved Gemini Robotics 2 can manage "whole-body motions," ranging from feet to fingertips.
Enhanced capabilities
Robots can now walk, stretch, and crouch
The upgraded model, Gemini Robotics 2, gives humanoid robots the ability to perform a wider range of actions. These include walking, crouching, stretching, and manipulating objects.
Google has shared videos showing Apptronik's Apollo 2 robot using this new capability to pick up a watering can or retrieve specific items from a shelf.
Improved manipulation
Gemini Robotics 2 also improves dexterity
The latest model also improves dexterity, letting robots control more complex five-fingered hands.
This means they can now do things like sealing a Ziploc bag, tying a trash bag, or unscrewing a lightbulb.
The update is seen as an important step toward enabling robots to perform more complex real-world tasks that require whole-body coordination.
Advanced reasoning
Update to Gemini Robotics ER
Along with the new model, Google DeepMind is also updating Gemini Robotics ER (embodied reasoning), a vision-language model that helps robots analyze their surroundings and process instructions.
The updated version, Gemini Robotics ER 2, is better at completing tasks over an extended period of time and "now understands when tasks begin and end."
This improvement allows multiple types of robots to work together on tasks.
Enhanced safety
Google's 'safest robotics model to date'
The updated model, Gemini Robotics ER 2, is also being touted as Google's "safest robotics model to date." It can better detect when humans are nearby and trigger safety tool calls.
The system can even bring the robot to a safe stop if someone approaches too closely.