Google announced a Robotics plugin for Antigravity on November 17, 2026, pairing Gemini 4 Hydrogen with tools for setting up supported robot arms, testing tasks in simulation and controlling physical hardware. Its pitch: go from a robot in a box to a working application in an afternoon. Source
The plugin’s repository warns that moving robots can injure people and damage equipment. It tells users to stay beside the robot, keep the workspace clear and keep a way to cut power within reach. Responsibility for safe operation stays with the user. This is supervised hardware, not a background browser tab. Source
The code hasn’t disappeared. Gemini writes it.
According to Google, Hydrogen takes a plain-language goal and a camera view, plans the task, writes and executes control code, then adjusts using feedback. The plugin bundles driver installation, camera discovery, arm calibration, object detection and grasp planning into that workflow. Source
Hydrogen itself isn’t making its debut here. Google’s announcement says the model was introduced the previous week; the new launch is the Robotics plugin that brings these tools together inside Antigravity. Source
Google describes one setup using an I2RT YAM arm and an external depth camera connected to a laptop. After software installation, calibration and test motions, the company says the arm was ready for tasks in about 10 minutes. It also describes onboarding examples with SO-ARM101 and FR3 Duo hardware. Those are named examples, not a promise that any arm with a USB cable will cooperate—or that every setup takes 10 minutes. Source
Give the snacks a logistics department
MuJoCo, the physics simulator, runs inside Antigravity. Google says developers can build scenes from descriptions, documents or workspace photos, then adjust shapes, positions and physical properties through chat. The intended loop is to model a real setup, test the controller in simulation and deploy it to the physical robot. The repository says the same controller code can run in simulation and on the arm. SourceSource
For its snack-sorting conveyor example, Google describes starting with a whiteboard sketch and images of the physical scene, building a MuJoCo simulation, testing the grasping policy and deploying it to the robot. Further conversational adjustments changed gripper force and sorting rules. Google says the process took about an hour end to end, producing an arm that reliably sorted snacks from the moving belt. Snack duty has acquired a development pipeline. Source
Google doesn’t give a success rate for the sorter or typical setup times across different hardware. Source
How to try it
Google’s launch instructions say to select gemini-4-hydrogen in Antigravity, open Customizations, and find Robotics on the Build With Google shelf. Antigravity requires a paid Google AI Pro or Ultra subscription. Source
The repository also documents GitHub installation, though that route requires installing the MuJoCo skills separately; the Marketplace version bundles them. And you don’t need to buy a robot to explore: the repository explicitly supports simulation-only use. Source
Comments