Google DeepMind Pushes Gemini AI to Power Humanoid Robots

Google DeepMind is expanding its artificial intelligence ambitions beyond software into physical machines, according to a report by Scientific American. The company is working to integrate its Gemini AI model into humanoid robots as part of a broader push into what the publication describes as "physical AI."

The initiative marks a shift for Gemini, which has primarily functioned as a digital assistant and language model, into a system meant to control robotic bodies operating in real-world settings. According to Scientific American, this effort is testing whether the intelligence embedded in Gemini can function reliably outside of controlled digital environments.

The report frames the project as an attempt to determine whether an AI model built for tasks such as conversation, reasoning and content generation can be adapted to guide physical machines through the complexities of the physical world, including navigation, manipulation and interaction with unpredictable surroundings.

Scientific American did not specify in the excerpt which robotics manufacturers or specific humanoid platforms are involved in the testing, nor did it detail a timeline for when Gemini-powered robots might be deployed commercially.

The move reflects a broader trend among AI developers to extend large language and multimodal models beyond screens and into embodied systems, a field often referred to as embodied AI or physical AI, where software-driven intelligence is paired with sensors, actuators and mechanical bodies to perform tasks in real environments.

Further details on the scope, partners and results of Google DeepMind's robotics testing were not fully outlined in the available source material.