Google DeepMind is pushing the boundaries of robotics with its latest innovation, Gemini Robotics 2. This AI system represents a significant leap from its predecessor, enabling robots to think, move, and interact autonomously in unpredictable environments. Unlike traditional robots that often rely on pre-programmed routines or remote control, Gemini Robotics 2 empowers robots to navigate challenges independently, bringing us closer to the sci-fi visions of robotic capabilities.
Can Robots Finally Handle Whole-Body Tasks?
Historically, Google’s robotics advancements have primarily focused on upper-body tasks, typically limited to actions performed on tabletops. With the introduction of Gemini Robotics 2, however, the entire humanoid robot is now under precise control, from its feet to its fingertips. A notable demonstration showcased Apptronik’s Apollo 2 robot executing a task by walking over to pick up a watering can and placing it into a bin on a bottom shelf with pinpoint accuracy.
Google DeepMind
This updated model boasts enhanced dexterity, enabling it to control a five-fingered robotic hand adeptly enough to tie knots, seal zip-lock bags, or remove light bulbs. Additionally, the system performs effortlessly with simpler two-fingered grippers for tasks such as packing and sorting.
In tandem with this advancement, Google has unveiled a new reasoning model, Gemini Robotics ER 2. Serving essentially as a project manager for the robot, it interprets instructions, monitors extended tasks, and facilitates coordination among multiple robots tackling the same project. Notably, an on-device version exists for robots operating without internet access, allowing them to adapt to new robot bodies swiftly—often in just a few hours and with as few as 200 examples.
What Makes These Robots Safer to Work With?
Safety has been a paramount focus in the development of Gemini Robotics 2. Google has introduced a new benchmark, ASIMOV-Agentic, which rigorously assesses robots’ ability to recognize when to decline hazardous actions or request human assistance. The Gemini Robotics ER 2 is equipped to detect when humans are too close, enabling it to halt to ensure safety. This model is already operational on Google AI Studio, while the others are gradually rolling out to early access partners.
This progressive leap in robotic technology paves the way for a future where robots can assist in more complex environments, broadening the scope of their applications in industries ranging from healthcare to logistics and beyond.
For more insights, you can read the full article here.
Image Credit: www.digitaltrends.com
