Genesis World is a simulation platform for physical AI development, formerly named Genesis, launched as an academic project in December 2024 and now officially supported by Genesis AI. It integrates a unified multiphysics engine, the photorealistic renderer Nyx, and the cross-platform compiler Quadrants behind a Python-style simulation interface, with the design goal of scaling from a single laptop core to data-center-class GPUs while remaining readable, extensible, and easy to embed in research code.
Purpose and Positioning Genesis World targets general robotics and embodied AI learning and can be used to build robot environments, machine learning pipelines, data generation, and agent simulation. Architecturally, the platform is divided into four layers: simulation interface, physics, rendering, and compiler.
Main Capabilities The simulation interface is the user-facing API, supporting asset parsing (URDF, MJCF, OBJ, GLB, USD, etc.), entity accessors, controllers, sensors, parallel and heterogeneous environments, and a built-in GUI. The physics layer is a unified multiphysics engine integrating Rigid, FEM, MPM, Particle (PBD / SPH), uipc, explicit couplers, and SAP, all sharing the same scene and the same state. The rendering layer provides three rendering paths that can be attached as camera sensors: Nyx (a self-developed renderer designed for robotics), Luisa (a DSL ray tracer), and Pyrender (a rasterizer). The compiler layer Quadrants can lower Python kernel code to CUDA, AMD ROCm, Apple Metal, Vulkan, x86, and ARM64, and hosts Genesis's automatic differentiation, GPU graph, and fastcache mechanisms.
Usage Before installation, PyTorch must first be installed according to the official instructions, and then Genesis can be installed via PyPI. To get the latest version, first run pip install --upgrade pip to update pip, then run the installation command; note that the package still needs to be updated manually to sync with the main branch. Users who wish to contribute are advised to install in editable mode: first ensure that genesis-world has been uninstalled, then clone the repository and install locally; after moving HEAD, it is recommended to systematically run pip install -e ".[dev]" to ensure that all dependencies and entry points remain up to date. Optional extras: Quadrants is already bundled with Genesis automatically, with no additional installation required; if you want to use the compiler outside Genesis, you can refer to the standalone wheel (pip install quadrants). uv is a fast Python package and project manager; you can use uv to get started quickly, then install PyTorch for the platform and run the examples. The repository provides runnable demos in three sections, corresponding to the layers of Genesis: physics (solvers and multi-solver coupling), rendering (in-repository camera setup and the Nyx walkthrough hosted at genesis-nyx), and simulation interface (sensors, GUI, controllers, parallel/heterogeneous environments, and tutorials). Most scripts can run end to end after pip install -e ".[dev]"; demos that depend on optional backends (such as IPC and Nyx examples) require installing additional dependencies as listed under "Optional extras."
Applicable Scenarios Suitable for robot environment construction, machine learning pipelines, data generation, agent simulation, and physical AI research requiring multiphysics coupling, photorealistic rendering, and cross-platform GPU acceleration.
Limitations and Notes The materials do not provide specific pricing, performance metrics, or platform compatibility details; some demos depend on optional backends and require additional installation. The project uses the Apache 2.0 license. Ways to contribute include submitting pull requests for new features or fixes, reporting defects through GitHub Issues, and proposing suggestions to improve usability; you can report problems or request features through GitHub Issues and participate in discussions or ask questions in GitHub Discussions.
