Google DeepMind Unveils Gemini Robotics 2 With Full-Body AI Control

Google DeepMind has introduced Gemini Robotics 2, a next-generation artificial intelligence system designed to provide humanoid robots with complete body control. Announced on 31 July, the new platform extends robot capabilities beyond upper-body movement, enabling humanoid machines to walk, crouch, manipulate objects, and collaborate with other robots. The release represents another step towards developing more capable and adaptable robotic systems for real-world applications.

From Upper-Body Movement to Whole-Body Control

The biggest advancement in Gemini Robotics 2 is its ability to coordinate the movements of an entire humanoid robot instead of focusing mainly on its arms and hands, as the first-generation model introduced in March did.

Google describes Gemini Robotics 2 as a vision-language-action model, combining visual understanding, natural language comprehension, and physical control to convert spoken or written instructions into coordinated actions.

During a demonstration, Apptronik’s Apollo 2 humanoid robot was asked to place a watering can on a lower shelf. The robot interpreted the instruction, walked to the table, picked up the watering can, and placed it in the requested location without additional manual programming.

Three AI Models Power the Platform

Google DeepMind has introduced three specialised AI models as part of the latest robotics platform.

Model	Primary Function
Gemini Robotics 2	A vision-language-action model that controls full-body movement and physical interaction.
Gemini Robotics ER 2	An embodied reasoning model that performs high-level planning, decision-making, and multi-robot coordination.
Gemini Robotics On-Device 2	A lightweight model that runs locally without internet connectivity and adapts quickly to different robot designs.

According to Google DeepMind, Gemini Robotics On-Device 2 can adapt to entirely new robot bodies using only a few hours of data and fewer than 200 examples, making deployment more efficient across a wide range of robotic platforms.

Improved Dexterity and Robot Collaboration

Beyond mobility, Gemini Robotics 2 significantly improves fine motor control. Working with Apptronik’s Apollo 2 robot and its five-fingered, 22-degree-of-freedom hand, the AI can perform delicate tasks such as tying knots and sealing ziplock bags.

Google DeepMind noted that while these capabilities represent meaningful progress, the precision and speed of the robotic hand have not yet reached human-level hand technology.

Another major enhancement is multi-robot collaboration. Robots powered by the platform can communicate through shared semantic understanding, enabling different machines to coordinate tasks, exchange contextual information, and complete workflows that would be difficult for a single robot to perform independently.

Carolina Parada, Head of Robotics at Google DeepMind, described the release as another milestone toward “physical AGI,” Google’s long-term vision of enabling robots to eventually perform a broad range of physical tasks that humans can carry out.

Safety Remains a Key Focus

Alongside Gemini Robotics 2, Google introduced ASIMOV-Agentic, a benchmark designed to evaluate the safety of layered AI systems that control robots.

The benchmark measures whether AI systems can recognise potentially harmful commands, avoid unsafe actions, detect nearby people, and stop appropriately before causing harm. By introducing a dedicated safety evaluation framework, Google aims to support the responsible deployment of increasingly capable robotic systems.

Industry Context

The launch comes during a period of rapid growth in AI-powered robotics. In related industry developments, the U.S. Federal Communications Commission (FCC) has proposed restrictions on future sales of certain Chinese-made robots over security concerns. If implemented, the proposal could affect the availability of some robot hardware in the United States, providing additional context to the evolving robotics landscape.

Key Highlights

Feature	Details
Developer	Google DeepMind
Platform	Gemini Robotics 2
Main Upgrade	Full-body humanoid control
Core Technology	Vision-language-action AI
AI Models	Gemini Robotics 2, Gemini Robotics ER 2, Gemini Robotics On-Device 2
Offline Support	Yes, via On-Device 2
Adaptation	Few hours of data and fewer than 200 examples
Dexterity	Knot tying and ziplock bag sealing
Multi-Robot Support	Shared semantic understanding and coordinated task execution
Safety Framework	ASIMOV-Agentic benchmark

Why This Matters

The launch of Gemini Robotics 2 demonstrates Google’s continued investment in embodied AI by extending robot capabilities from upper-body movement to coordinated full-body actions. Features such as on-device intelligence, improved dexterity, and collaborative robotics expand the potential for applications across manufacturing, logistics, healthcare, research, and service industries.

The introduction of the ASIMOV-Agentic benchmark also highlights the growing importance of evaluating safety as AI-powered robots become more autonomous and capable of operating in environments shared with people.

Conclusion

The release of Gemini Robotics 2 marks a significant advancement in humanoid robotics by combining full-body movement, natural language understanding, embodied reasoning, fine motor control, and multi-robot collaboration within a unified AI platform. Together with the ASIMOV-Agentic safety benchmark, the latest platform reflects Google’s focus on developing capable and responsible robotic systems. As embodied AI continues to evolve, innovations like Gemini Robotics 2 are expected to expand the practical role of humanoid robots across a wide range of real-world applications while maintaining a strong emphasis on safety and collaboration.