Google DeepMind Aims for Robust Physical AI with General-Purpose Intelligence
Google DeepMind is working on a project to create an AI brain that can jump between robot bodies. This 'physical AI' will have general-purpose intelligence, allowing it to inhabit different machines and perform various tasks.
The company demonstrated its latest attempt at this technology with Gemini Robotics 2 in late July. This software can control whole-body movement, including legs, torso, arms, and fingers, across a range of robots from two-armed research platforms to humanoids like Apptronik's Apollo 2.
DeepMind is using machine learning to enable its AI brain to learn from one environment and apply that knowledge to another. This approach has been successful in robotics, as seen with the AlphaGo system, which mastered the complex board game Go through reinforcement learning.
The team at Google DeepMind is now focusing on 'dexterous manipulation,' allowing robots to perform everyday tasks like folding laundry or making scrambled eggs. They believe that by understanding how humans interact with their environment, they can create a more versatile and capable physical AI.