Google DeepMind has unveiled the latest iteration of its Gemini Robotics AI model, Gemini Robotics 2, which now boasts the capability to "control entire humanoid robots." This marks a significant leap from its predecessor, which focused primarily on upper-body control, as Gemini Robotics 2 extends its command to "whole-body motions," encompassing everything from a robot's feet to its fingertips, as announced on Thursday.
This enhanced model empowers humanoid robots to execute a broader spectrum of actions, including walking, crouching, stretching, and intricate object manipulation. Demonstrations shared by Google showcase Apptronik’s Apollo 2 robot seamlessly bending to retrieve a watering can and methodically identifying and selecting specific items from a shelf.
While Google DeepMind acknowledges that its robots "have more to advance in movement speed," the company underscores that this update represents "an important step towards the skills needed to complete more complex, real-world tasks that require whole-body coordination."
Further improving capabilities, Gemini Robotics 2 offers superior dexterity, enabling the control of more sophisticated, five-fingered hands. This advancement allows robots to perform delicate tasks such as sealing a Ziploc bag, expertly tying a trash bag, or precisely unscrewing a lightbulb.
Concurrently, Google DeepMind has also updated its Gemini Robotics ER (embodied reasoning) model. This vision-language model is crucial for robots to interpret their surroundings, process instructions, and manage multi-step operations. The new Gemini Robotics ER 2 excels at sustaining tasks over extended periods and "now understands when tasks begin and end," enhancing its temporal awareness.
This update also facilitates collaborative work among diverse robot types. A compelling video illustrates Apollo 2 directing Google’s dual-arm robot to place tools into a bin, demonstrating coordinated efforts during a garage cleaning task.
Emphasizing safety, Google DeepMind states that Gemini Robotics ER 2 is its "safest robotics model to date." It features improved detection of nearby humans, can trigger safety protocols, and will bring the robot to a safe stop if someone approaches too closely.
In parallel developments, Google DeepMind has refined its Gemini Robotics On-Device Model, which operates locally on a robot without requiring an internet connection. This model now adapts more swiftly to new robotic embodiments, including those possessing "drastically different shapes, sensors and degrees of freedom."
The Editorial Staff at AIChief is a team of professional content writers with extensive experience in AI and marketing. Founded in 2025, AIChief has quickly grown into the largest free AI resource hub in the industry.