Find your next open-source project

Search by name, purpose, or keyword — or paste a GitHub repository URL

google-deepmind/robotics-antigravity-plugin

Hardware & IoT

Agent skills that let Gemini-based coding agents build MuJoCo simulations and control real robot arms.

122
11
Python
Apache-2.0
2 days old

What it is

This is a plugin for Google Antigravity that packages robotics "skills" for an agentic workflow. The agent writes a controller, tests it in MuJoCo simulation, and is meant to run the same code on a real robot arm, with simulation-only work also possible. It targets people who want spatial reasoning, multimodal perception, and code generation inside the robot loop rather than building those pieces by hand.

Example prompts cover simulation-only tasks (a pelican riding a bicycle, a grocery sorting task with a LeRobot SO-101 arm, a bimanual Franka coffee-shop scene) and real-robot tasks such as bringing up an SO-101 or I2RT YAM arm to pick up objects and record video. More background on the underlying model is at Gemini Robotics.

Install and first run

  1. Install the latest Google Antigravity.
  2. Install the plugin from GitHub with the Antigravity CLI:
agy plugin install https://github.com/google-deepmind/robotics-antigravity-plugin

If you prefer a manual copy, clone the repository, place plugins/* under ~/.gemini/config/plugins/, and restart Antigravity:

git clone https://github.com/google-deepmind/robotics-antigravity-plugin.git
mkdir -p ~/.gemini/config/plugins
cp -r robotics-antigravity-plugin/plugins/* ~/.gemini/config/plugins/

Individual skills can also be added with npx skills:

npx skills add google-deepmind/robotics-antigravity-plugin
  1. If the plugin was installed from this GitHub repository rather than the Antigravity Marketplace, the seven MuJoCo API skills are not bundled. Install them from the canonical MuJoCo skills directory:
npx skills add google-deepmind/mujoco
  1. For skills that call Gemini perception models, set GEMINI_API_KEY in your environment before running.

The expected result is an agent session in which you can ask for a simulation or robot task as in the README examples; from there you run the generated controller and inspect the trajectories or recorded video.

Requirements and limits

  • The Antigravity route requires Google Antigravity; the repository is structured around its plugin discovery manifest and plugins/robotics directory.
  • Real-robot use assumes compatible hardware such as an SO-101 or I2RT YAM Pro arm and whatever bring-up skills the plugin provides; simulation-only use needs no robot.
  • Some skills download third-party software or robot models, or call third-party services such as the Gemini API at run time. Those components carry their own licenses and terms, separate from the repository's Apache-2.0 license.

Safety

The README warns that a moving robot can injure people and damage equipment. Keep the robot in view, clear the workspace, and keep a way to cut power within reach; operating the robot safely is the user's responsibility.

HotRepo's take

The repository recorded 109 stars and 10 forks on 2026-11-19, and appeared on the Python new-project list for that date with a gain of 5 stars over roughly one hour. The sample is short and there is no broader trend to read into. What stands out technically is the packaging choice: rather than shipping a runtime, it distributes an agent plugin plus skills, with MuJoCo API skills deliberately delegated to the MuJoCo project unless the plugin comes from the Marketplace.

Written by AI from the project's README, updated . The project's own docs are the source of truth.

Hacker News

Show HN: Try Vibe Robotics! (IRL or MuJoCo SIM)
▲ 11 · 0 comments · November 18, 2026

You might also like

  1. google-deepmind/mujocoGeneral-purpose physics engine for simulating articulated robots and structures, used by robotics, biomechanics, and machine learning researchers.C++16,013
  2. mit-han-lab/pearPost-training engine that trains reasoning models to control real robots via reinforcement learning in simulation.Python103
  3. rerun-io/rerunOpen-source SDK and viewer for logging, visualizing, and querying multimodal robotics and sensor data over time.Rust11,710
  4. intrinsic-ai/intrinsic-coreA local runtime, SDK, and real-time control framework for building industrial robotics applications on ROS 2.C++754
  5. NVIDIA/skillsA catalog of NVIDIA-verified Agent Skills that teach Claude Code, Codex and other coding agents NVIDIA workflows.Python3,988