Genesis is a physics simulation platform designed for general-purpose robotics, embodied AI, and physical AI applications. It aims to lower the barrier to entry for robotics research by providing a user-friendly, high-fidelity simulation environment that automates data generation.

How It Works

Genesis is built upon a universal physics engine that integrates various physics solvers (MPM, SPH, FEM, PBD, Stable Fluid) into a unified framework. This core engine supports a wide range of material models and robot types, and is enhanced by a generative agent framework for automated data generation. The platform emphasizes speed, photo-realism via native ray-tracing, and differentiability across its solvers.

Quick Start & Requirements

• Install via pip: pip install genesis-world (Python >= 3.10, < 3.13) or from source.
• Requires PyTorch installation.
• Docker image available for GPU-accelerated ray-tracing.
• Documentation available in English, Chinese, and Japanese.

Highlighted Details

• Achieves over 43 million FPS on a single RTX 4090 for robotic arm simulation.
• Cross-platform support (Linux, macOS, Windows) with multiple compute backends (CPU, Nvidia/AMD GPUs, Apple Metal).
• Integrates diverse physics solvers and material models, including rigid bodies, liquids, gases, and deformable objects.
• Supports various robot formats (URDF, MJCF) and 3D model formats.
• Features native ray-tracing for photo-realistic rendering.
• Differentiability is supported for MPM and Tool solvers, with plans for others.

Maintenance & Community

• Active development with recent releases (v0.2.1 on Jan 8, 2025).
• Discord and WeChat groups available for community interaction.
• Open to community contributions via pull requests and bug reports.

Licensing & Compatibility

• Licensed under Apache 2.0.
• Compatible with commercial use and closed-source linking.

Limitations & Caveats

The generative framework is modular and being gradually rolled out; full access to generative features will be available in the near future. Differentiability for solvers other than MPM and Tool is planned for future versions.