Google DeepMind: Gemini Robotics 2 Controls Robots

Google DeepMind: Gemini Robotics 2 Controls Robots

Google DeepMind announces that the latest version of its AI model Gemini Robotics can "control humanoid robots in their entirety." While the previous model focused on controlling the upper body of a humanoid robot, Gemini Robotics 2 now supports "full-body movements," from feet to fingers, according to an announcement made on Thursday.

The new model will enable humanoid robots to perform a wider range of actions, allowing them to walk, squat, stretch, and manipulate objects. Videos shared by Google demonstrate how the Apollo 2 robot from Apptronik can bend down to pick up a watering can, as well as find and retrieve specific items from a shelf.

Although Google DeepMind notes that its robots "still have progress to make in terms of movement speed," it adds that this update "is a significant step toward the skills needed to perform more complex real-world tasks that require full-body coordination."

Additionally, Gemini Robotics 2 supports better dexterity, as it can now control more complex five-fingered hands. This enables robots to perform tasks such as sealing a Ziploc bag, tying a trash bag, or unscrewing a light bulb.

Google DeepMind is also updating Gemini Robotics ER (Embodied Reasoning), a vision-language model that helps robots analyze their environment, process instructions, and carry out multi-step tasks. Gemini Robotics ER 2 is more efficient at completing tasks over extended periods and "now understands when tasks start and end."

Google DeepMind states that this update also allows multiple robots of different types to work together and accomplish tasks, with a video showing how Apollo 2 instructs Google's dual-arm robot to place tools in a box while cleaning the garage.

The company specifies that Gemini Robotics ER 2 is its "safest robotics model to date," as it can "better detect when humans are nearby, trigger safety calls, and bring the robot to a safe stop if someone approaches too closely."

Meanwhile, Google DeepMind has made improvements to its Gemini Robotics On-Device Model, which can operate locally on a robot without an Internet connection. This model can now adapt more quickly to new embodiments, including those with "radically different shapes, sensors, and degrees of freedom."