Google DeepMind unveils Gemini Robotics 2 for adaptable robots

Google DeepMind, Alphabet's AI subsidiary, announced Gemini Robotics 2 in a blog post, describing it as "the intelligence layer powering the next generation of truly adaptable robots." Unlike the predecessor Gemini Robotics from 2025, which focused on upper-body tasks such as controlling table-mounted gripper arms, the second generation coordinates complex movements involving the entire body of humanoid robots. In a demo video, a humanoid robot from Apptronik picks up a watering can from the floor, bending and gripping it smoothly, albeit slowly. The company also touts a "new dimension of motor dexterity" for both human-like hands and grippers; in the demo, a robot ties garbage bag ties, a task that many small steps for humans. Additionally, robots of different types and manufacturers can collaborate with Gemini Robotics 2, sharing insights through machine learning to solve tasks a single robot couldn't handle. The product family includes three models: Gemini Robotics 2, the most advanced Vision-Language-Action (VLA) model converting visual and language commands into motor actions; Gemini Robotics ER 2, an Embodied-Reasoning model for perceiving and interacting with the physical world; and Gemini Robotics On-Device 2, a vision-language model running locally on robots without internet. The company aims to move closer to practical, human-like robots that assist with daily chores.