Google's robotics technology makes a major breakthrough! Gemini Robotics 2 achieves full-body control and enters the physical world

Alphabet's Google DeepMind officially launched its next-generation humanoid artificial intelligence model, Gemini Robotics 2, this week. This breakthrough development enables humanoid robots to achieve full-body motion coordination from head to toe, marking a critical step for the tech giant in extending artificial intelligence capabilities into the physical world.

The biggest technical highlight of Gemini Robotics 2 is its expansion of control over humanoid robots from the previously limited upper body to overall full-body control. This means that while executing complex reasoning tasks, the robot can simultaneously perform high-difficulty actions such as walking, squatting, and dexterously manipulating objects.

In a recent demonstration to the media, the model successfully controlled Apptronik's Apollo humanoid robot to autonomously navigate around obstacles, cross a room, accurately pick up a watering can, and place it on a lower shelf, showcasing a high level of autonomous coordination.

At the same time, Google DeepMind also unveiled two supporting models, Gemini Robotics ER 2 and On-Device 2. These two new products can operate independently or work together, aiming to support robots in planning multi-step tasks and coordinating multiple robots to perform operations, further enhancing its robotic AI software ecosystem.

Although this release demonstrates Google's strong capabilities in the field of physical artificial intelligence, the company maintains a calm assessment of the current technological state. Carolina Parada, Vice President of the Google DeepMind Robotics Division, emphasized that the company's long-term goal is to bring AI into the physical world and build a universal and efficient layer of robotic intelligence.

However, Kanishka Rao, Director of the Robotics Division, also admitted that pursuing true robotic flexibility remains a long-term challenge.

Currently, the robot's motion process still appears relatively slow and deliberate, as the robot must stop to process and think about judgments that humans can complete intuitively before executing actions. This provides a more cautious perspective on the commercialization process of humanoid robots.

The release of this model not only represents a continued deepening of Google's strategy in the robotics field but also reignites its long-standing ambitions in robotics that have spanned over a decade. Since the launch of the first-generation Gemini Robotics in 2025, Google has reestablished its development direction of converting language and visual information into robotic actions.

Looking back, Google's layout in the robotics field has not been smooth sailing. After experiencing intensive acquisitions in the early 2010s followed by business contractions, and even closing the Everyday Robots division in 2023, this frequent and high-intensity release of technology indicates Google's renewed investment intensity in this competition With the expansion of the artificial intelligence front, Google is facing intense external competition in the field of robotics. OpenAI is currently actively exploring the integration of vision, language, and action into a general foundational model, while NVIDIA is assisting developers in training AI-driven robots by providing powerful software toolchains.

Under the pressure of technological competition and market expectations, Google is attempting to occupy a technological high ground in the future humanoid robot industry with its Gemini Robotics series of models