Google Unveils Gemini Robotics 2 AI Model
Google DeepMind has introduced Gemini Robotics 2, its latest AI foundation model designed to give humanoid robots advanced whole-body intelligence. The new system expands beyond upper-body manipulation by enabling robots to walk, crouch, reach, balance, and perform coordinated full-body movements while handling complex physical tasks. Consequently, the release marks a significant step toward more capable general-purpose robots for homes, factories, and warehouses.
Alongside Gemini Robotics 2, Google also launched Gemini Robotics ER 2 (Embodied Reasoning), an upgraded reasoning model that improves task planning, video understanding, and multi-robot collaboration. Moreover, the two models work together to help robots understand their surroundings, plan actions, and complete multi-step tasks with greater reliability.
Whole-Body Intelligence Enhances Robot Capabilities
Unlike previous versions that primarily focused on arm manipulation, Gemini Robotics 2 controls an entire humanoid robot from head to toe. Therefore, robots can coordinate walking, bending, stretching, and precise hand movements while adapting to changing environments.
Google demonstrated robots performing a variety of real-world tasks, including organizing shelves, picking specific objects, watering plants, inserting tapes into a stereo, tightening light bulbs, and tying trash bags. Furthermore, the model supports five-finger dexterity, allowing robots to handle delicate manipulation tasks that require coordinated finger movements.
The company also reported improvements in robotic perception and task execution. As a result, robots can better understand instructions, monitor task progress, and determine when an activity has been completed before moving to the next objective.
Gemini Robotics ER 2 Improves Reasoning and Safety
Google DeepMind introduced Gemini Robotics ER 2 to strengthen high-level reasoning for robotic systems. The upgraded model enhances video understanding, orchestrates tools, and enables multiple robots to collaborate on shared tasks. Consequently, robots can execute longer workflows while maintaining awareness of task progress and environmental changes.
The company also expanded its safety framework. New capabilities include improved human proximity detection, automatic safety stops, and the ASIMOV-Agentic benchmark, which evaluates the safety of AI-powered robotic behavior before deployment. These additions aim to support responsible development as robots become increasingly autonomous.
Advancing Google’s Vision for Physical AI
Gemini Robotics 2 reflects Google’s broader ambition to extend Gemini beyond digital assistants into the physical world. Therefore, the platform is designed to support multiple robot types rather than a single humanoid platform, allowing manufacturers to adapt the technology to different hardware configurations.
The announcement also intensifies competition in embodied AI, where Google DeepMind joins companies such as NVIDIA, Boston Dynamics, and other robotics developers working toward more capable humanoid systems. As enterprise demand for intelligent automation grows, Google expects Gemini Robotics 2 to accelerate the deployment of robots that can safely perform increasingly sophisticated real-world tasks.
