The Google DeepMind lab unveiled a breakthrough in robotics on June 25, 2025 – a new line of AI models called Gemini Robotics On-Device. This announcement, as reported by leading tech publications including Business Insider, is a key step towards creating truly autonomous, adaptive, and intelligent robots. The main feature of the new models is their deep optimization for efficient operation directly on the robots own hardware ("on-device" or "edge inference"), without the need for constant communication with remote cloud servers. This solves one of the fundamental problems in modern robotics – latency in data transmission and processing, which critically hinders robots from reacting quickly and adequately to dynamic changes in their environment. The Gemini Robotics On-Device models are specifically trained to perform a wide range of robotic tasks: from complex visual perception and building 3D maps of the surroundings to high-precision object manipulation and understanding natural language commands. Running AI computations locally also significantly increases system reliability (the robot can continue to operate even without an internet connection) and data security, as sensitive visual and other information does not leave the device. This technology paves the way for a new generation of robots in fields such as smart manufacturing (flexible assembly), warehouse automation, last-mile logistics, and even for creating personal assistant robots in healthcare and domestic settings. By providing a powerful "brain" that operates directly "on-board," Google DeepMind is significantly accelerating the transition from rigidly programmed machines to truly adaptive and autonomous robotic systems.
Google DeepMind Unveils Gemini Robotics – AI Models for Autonomous Robots
