Google DeepMind has unveiled Gemini Robotics 2, an upgraded AI platform designed to control a wide variety of robots, including humanoid machines. In a demonstration video, the robots perform practical tasks such as cleaning up trash, picking up watering cans, inserting a cassette into a boombox, screwing in a lightbulb, and tying a garbage bag, highlighting the platform's advanced whole-body control and real-world capabilities.

The original Gemini Robotics platform enabled robotic arms and hands to perform precise tasks such as sealing storage bags and folding origami. With Gemini Robotics 2, Google DeepMind expands those capabilities by introducing intelligent whole-body control, allowing robots to coordinate full-body movements for more complex real-world tasks.

Figure 1. Gemini Robotics 2.

While some remain skeptical, Google says the demonstration video features fully autonomous robots operating in real time. The company contrasts this with earlier demonstrations of Elon Musk's Optimus robots, where teleoperators were later revealed to have remotely controlled the machines during a high-profile showcase. Figure 1 shows Gemini Robotics 2.

Google's Gemini Robotics 2 combines multiple AI models to give robots a multimodal understanding of their environment. A vision-language model interprets surroundings, while two vision-language-action models coordinate full-body and hand movements. Although the robots are not yet general-purpose, they were trained for specific tasks using human teleoperation, video demonstrations, and simulations. Google DeepMind describes the platform as another step toward achieving "physical AGI," with the long-term goal of enabling robots to perform any task a human can.

Because AI-powered robots can make mistakes, safety remains a major concern, especially when operating large, human-sized machines. Google says Gemini Robotics 2 uses a multi-layered safety framework with guardrails built into every AI model [1]. The company has also introduced the ASIMOV-Agentic benchmark to evaluate whether a robot's actions could lead to harmful outcomes. According to Google DeepMind, improving robot safety is essential as robots are deployed in more complex and unpredictable real-world environments.

Google's humanoid robotics technology is still in its early stages, with no plans for consumer-ready robots in the near future. Meanwhile, Tesla CEO Elon Musk has made ambitious predictions for the Optimus robot, including massive production and trillions of dollars in potential sales. However, several of those projected deployment milestones have not been achieved, underscoring the gap between long-term goals and current progress.

References:

1. https://www.engadget.com/2227268/google-gemini-robotics-2-platform-intelligent-whole-body-control/

Cite this article:

Janani R (2026), Google Unveils Gemini Robotics 2 With Whole-Body AI Control, AnaTechMaz, pp.394.