Skip to content

👋 Welcome to EmbodiedGen V2

🌐 Project Page GitHub 📄 arXiv 🤗 Dataset 🎥 Video

EmbodiedGen V2: An Agentic, Simulation-Ready 3D World Engine for Embodied AI.

From intent to executable 3D worlds. EmbodiedGen compiles language, images, and edit commands into simulation-ready 3D worlds — physically plausible assets, large-scale scenes, and task-driven interactive environments, deployable across every major robotics simulator.

EmbodiedGen V2 Overview

What can it do?

  • 🧱 Generate — Sim-Ready Assets

    Turn a single image or a text prompt into a simulation-ready asset: metric geometry, collision proxy, and physical properties.

    → Image-to-3D · Text-to-3D · Texture Generation · Soft-Body Simulation · Affordance

  • 🏠 Scale — Large-Scale Scenes

    Multi-room, navigable, instance-editable houses and photo-realistic 3DGS backgrounds at controllable complexity.

    → Room Generation · 3D Scenes

  • 🌍 Compose — Task-Driven Worlds

    Natural-language tasks become physically stable, directly loadable interactive layouts with robots and objects.

    → Layout Generation · Digital Twin Creation

  • 💬 Edit — 3D Vibe Coding

    Build and edit deployable world states through dialogue — every instruction a bounded, physics-validated skill call.

    → 3D Vibe Coding

  • 📦 Export — Any Simulators

    One standardized layout loads consistently across SAPIEN, Isaac Sim, Isaac Gym, MuJoCo, Genesis, and PyBullet.

    → Any Simulators

  • 🤖 Train — Robot Learning

    Generated worlds double as online training environments: parallel gym envs, trajectory recording, grasp evaluation.

    → Robot Learning

Start with the Installation Guide, then follow the Tutorials. Online demos live in Services.