GOOGLE DEEPMIND UNVEILS GEMINI ROBOTICS 2 WITH WHOLE-BODY CONTROL

Google DeepMind announced Gemini Robotics 2, a new artificial intelligence platform capable of controlling humanoid robots to perform a range of household tasks. The company released video footage showing robots cleaning up trash, picking up watering cans, putting tape into a boombox, screwing in a lightbulb and tying up a rubbish bag. Google stated the video features "real-time footage" of "fully autonomous" robots operating without human control during task execution. The update brings "intelligent whole-body control" to robotic systems, expanding capabilities beyond previous versions that relied primarily on robotic arms and hands.

The Gemini Robotics 2 platform combines several artificial intelligence models to give robots multimodal understanding of their surroundings. These include a vision language model for understanding and two vision language action models that control full-body and hand movement. Google DeepMind trained the model to perform each task shown in the video using a combination of human teleoperation, video examples and simulations. The company emphasised this represents progress towards what it calls "physical AGI" — the ability for a robot to perform tasks a human can accomplish.

Google acknowledged safety considerations in deploying such systems, stating it is taking a multi-layered approach to ensure future robots operate safely. Each model layer includes guardrails, and the company introduced a new benchmark called ASIMOV-Agentic designed to detect whether a command could lead to harmful outcomes. The company noted the robots are not general-purpose devices capable of performing any task, but rather systems specifically trained for the tasks demonstrated.