genesis-embodied-ai/genesis-world

Genesis is a physics platform designed for general-purpose robotics, embodied AI, and "physical AI", research where software agents have to interact with a simulated physical world before being deployed on real robots. Training those agents needs lots of realistic simulation, and existing simulators are often slow, hard to install, or limited to one kind of object. Genesis bundles several pieces into one package: a universal physics engine rebuilt from the ground up, a lightweight Python-friendly robotics simulation platform, a photo-realistic rendering system, and a "generative data engine" that aims to turn natural-language descriptions into useful simulation data. The way it works is that Genesis integrates several physics solvers, for rigid bodies, particle-based MPM and SPH for materials, finite-element FEM, position-based dynamics, and a stable-fluid solver, under one unified framework. That single framework can simulate rigid bodies, liquids, gases, deformable objects, thin shells, and granular materials, and couple them together. Robots are loaded from standard description formats including MJCF XML and URDF, plus mesh formats like obj, glb, ply, and stl. Genesis advertises very high simulation speed (over 43 million frames per second on a single RTX 4090 GPU when running a Franka robotic arm in their benchmark, they describe this as roughly 430,000 times faster than real time). It runs on Linux, macOS, and Windows, with support for CPU, Nvidia and AMD GPUs, and Apple Metal. Some solvers are also differentiable, which means gradients can flow back through the physics, useful for learning-based control. A native ray-traced renderer provides photorealistic visuals. Installation is via pip from PyPI, with a Docker option available, and Python 3.10 or newer is required. You would use Genesis if you are training or evaluating robotics policies, doing research in embodied AI, or generating synthetic data for robotic arms, legged robots, drones, or soft robots, and want a single Python-first toolkit that covers the physics, the rendering, and the data side. The repository is written primarily in Python.