Google DeepMind has introduced a next-generation artificial intelligence (AI) model that extends the.. - 매일경제

Google unveiled its next-generation AI model 'Gemini Robotics 2' on the 30th (local time). It is the follow-up model of the first generation introduced in March, and the biggest feature is that the control range, which was centered on the upper body, has been expanded to the entire body. Gemini Robotics is not a robot itself, but a "vision, language, and behavior (VLA) model that understands human voices, writings, and camera images and turns them into real movements. It also features that one AI can be applied to various humanoids and industrial robots.

The second generation can understand user instructions and perform tasks using the whole body, such as walking on their own, picking up items, leaning down and placing them on a shelf. In addition, delicate work is possible, such as tying a plastic bag or sealing a zipper bag with a robot hand equipped with five fingers and 22 joints. It also supports precision assembly and packaging using industrial tongs grippers. However, Google DeepMind said, "We are continuing to improve precision and speed because it has not yet reached human-level hand technology."

Google evaluated the model as an important milestone toward universal physical AI. If the first generation demonstrated the ability to control the robot's arms and hands, the second generation has evolved into a stage where movement, manipulation, reasoning, and collaboration are integrated into a single AI system to perform complex tasks in a real environment. Google said, "The goal is to build a general-purpose physical AI that solves complex problems in reality with people beyond single-task automation."

[Silicon Valley correspondent Wonho-seop]