Gemini Robotics 2 represents Google’s vision of robots powered not by rigid programming, but by adaptable intelligence capable of reasoning, learning, collaborating, and safely operating alongside humans in the physical world. (Source: Image by RR)

Multi-Robot Collaboration Marks a New Phase of Intelligent Automation

Google has unveiled Gemini Robotics 2, the next generation of its embodied AI platform designed to bring advanced intelligence to a wide range of robotic systems. Unlike traditional industrial robots that perform narrowly programmed tasks, Gemini Robotics 2 enables robots to reason through complex movements, adapt to unfamiliar environments, and perform coordinated actions involving their entire bodies—from walking and balancing to manipulating objects with human-like dexterity. The platform is built around three complementary AI models covering action, reasoning, and on-device intelligence.

A major advancement in Gemini Robotics 2 is its expansion from tabletop manipulation into full-body control. Humanoid robots can now walk, crouch, bend, reach, carry objects, and navigate real-world spaces while executing multi-step tasks. Google, as noted in a article at deepmind.google, also demonstrated substantial improvements in robotic dexterity, enabling both multi-fingered robotic hands and traditional grippers to perform increasingly delicate operations such as tying knots, sealing bags, organizing tools and handling complex assembly tasks with greater precision.

Beyond physical movement, Google introduced significant advances in embodied reasoning. Gemini Robotics ER 2 acts as a high-level planning system that allows robots to understand spoken instructions, communicate with people, plan workflows lasting several minutes, recover from mistakes, and even coordinate with other robots to complete shared objectives. Complementing this capability, Gemini Robotics On-Device 2 allows many of these functions to operate locally without relying on cloud connectivity, enabling robots to adapt to entirely new hardware platforms after only a few hours of additional training data.

Safety remains a central focus of the platform. Google introduced new benchmarks for uncertainty awareness, human proximity detection, and collaborative safety that allow robots to recognize when they should slow down, stop, or request human assistance. Together, these capabilities represent another step toward Google’s long-term vision of general-purpose physical AI—robots capable of working alongside humans across homes, factories, healthcare, logistics and countless other real-world environments.

read more at deepmind.google