© ROOT-NATION.com - Use of content is permitted with a backlink.
Google’s DeepMind division has brought fantastical robots closer to reality, which is truly awe-inspiring. Building on the first version of the Gemini Robotics model, the company has unveiled Gemini Robotics 2 – an AI system designed to help robots analyze their surroundings, navigate, and even collaborate with one another. Unlike most modern robotic systems, which rely on pre-programmed algorithms or remote control by an operator, this new solution allows robots to find their own way out of unpredictable situations.
See also: AERONAUT – everything that flies above the ground: aviation, UAVs and drones, rockets, and space
Are robots finally capable of performing tasks that engage their entire bodies? Until now, Google’s robotic models have primarily performed tasks on tabletops, using only the upper half of the robot’s body. Gemini Robotics 2 changes this approach by taking control of the entire humanoid device – from the feet to the fingertips. During one demonstration, Apptronik’s Apollo 2 robot was tasked with placing a watering can on the bottom shelf of a cabinet. The robot walked over, picked up the watering can, crossed the room, and placed it exactly where it was supposed to go.
This model also demonstrates significantly higher precision of movement. It is capable of controlling a five-fingered robotic hand with enough dexterity to tie a knot, close a zipper, or unscrew a light bulb, and works just as seamlessly with simple two-fingered grippers during packing or sorting.
Along with this, Google is introducing an analytical model called Gemini Robotics ER 2, which essentially acts as the robot’s project manager. It breaks instructions down into smaller steps, monitors the execution of tasks lasting several minutes, and can even coordinate the collaborative work of multiple robots on a single mission. In addition, there is an autonomous version designed for robots without internet access, which is capable of adapting to a completely new robot body in just 3–4 hours using only 200 examples.
Safety has been a top priority this time around. Google has created a new testing standard, ASIMOV-Agentic, to verify whether robots understand when to refrain from risky actions or, instead, ask a human for help. Gemini Robotics ER 2 is also capable of sensing when someone gets too close and safely stopping its operation. Gemini Robotics ER 2 is already available in Google AI Studio, while the rest of the models are currently being rolled out to early-access partners.
Read also: AI has broken free… again: Claude crashed the systems of three real companies during testing