DistantNews
Support us
🇰🇷 South Korea /Technology

Google DeepMind Unveils 'Universal Brain' AI for Robots Capable of Complex Tasks

From Hankyoreh · () Korean

Translated from Korean, summarized and contextualized by DistantNews.

At a glance

News Named sources Context piece
  • Google DeepMind has unveiled three new AI models for adaptive robot control, aiming to create a "universal brain" for robots.
  • These models enable robots to learn, reason, and perform tasks in unpredictable environments using full-body control.
  • While demonstrating impressive capabilities, the success rates for complex tasks like manipulating objects and assembling items remain a challenge.

Google DeepMind has introduced a suite of three advanced AI models designed to power adaptive robot control, heralding a new era for robotics. These models, collectively referred to as Gemini Robotics 2, aim to provide a "universal brain" capable of operating any robot, regardless of its specific hardware. The system allows robots to learn and reason in unpredictable environments, executing tasks through precise full-body movements, from fingertips to toes.

We hope Gemini becomes the AI operating system for robots, like Android is for smartphones.

— Demis HassabisGoogle DeepMind CEO Demis Hassabis expressed his vision for the Gemini models in robotics.

In a demonstration, a humanoid robot named Apollo 2, equipped with a robotic arm, responded to verbal commands. It successfully gathered items for a simulated sports outing, even adapting when a developer playfully moved a target bag. The AI system, comprising a visual-language model (VLM) for reasoning and a visual-language-action model (VLA) for control, translated abstract instructions into complex physical actions. This integration allows robots to perform tasks like tying knots, changing light bulbs, and assembling items.

Beyond individual robot control, one of the new models, Gemini Robotics ER2, also supports multi-robot collaboration. This feature enables different types of robots, humanoid, quadruped, or dual-arm, to communicate and coordinate on complex tasks that a single robot cannot accomplish alone. Google DeepMind envisions this as a significant step towards robots seamlessly integrating into daily life, assisting humans in various capacities.

We have dreamed of robots that can naturally integrate into our lives and help us for decades. That vision is now becoming a reality.

— Carolina ParadaCarolina Parada, head of robotics at Google DeepMind, shared her perspective on the future of robotics.

Despite the advancements, Google DeepMind acknowledges that challenges remain. Success rates for tasks requiring fine motor skills, such as screwing in a light bulb or sealing a bag, are still below ideal levels. Manipulating objects at lower positions, like those on the floor, also proved more difficult, indicating that full-body control in varied physical contexts requires further refinement. The company is focused on improving the speed and precision of robotic movements to bridge the gap between current capabilities and their long-term vision.

Fine manipulation with five fingers is still a difficult task.

— Google DeepMindGoogle DeepMind acknowledged the challenges in achieving precise robotic dexterity.
DistantNews Editorial

Originally published by Hankyoreh in Korean. Translated, summarized, and contextualized by our editorial team with added local perspective. Read our editorial standards.

More Stories

From Our Blog