Genesis

Universal physics engine and simulation platform for robotics and embodied AI at 430,000x real-time speed.

GitHub Open repository

29K stars 2.7K forks Apache-2.0 license 2026-06-04 verified

bash

$pip install genesis-world

Open source

Overview

What is Genesis?

Genesis is a universal physics platform re-built from the ground up for general-purpose robotics, embodied AI, and physical AI applications. It integrates a high-performance physics engine, photo-realistic rendering, and a generative data engine into one Pythonic framework. Supported by a generative agent framework, Genesis can transform natural language prompts into multimodal training data at unprecedented speed.

Unified multi-solver physics engine

Genesis integrates rigid body, MPM, SPH, FEM, PBD, and fluid solvers into a single framework, enabling coupled simulation of diverse materials.

Real-world tasks often involve multiple physical phenomena — a robot arm picking up a wet cloth, for example. Genesis simulates the whole scene without switching engines.

430,000x real-time simulation speed

Simulating a Franka robotic arm at over 43 million FPS on a single RTX 4090 — 430,000 times faster than real-time.

This speed makes it practical to generate massive synthetic datasets for training robot policies in hours instead of months.

Generative data engine from natural language

A generative agent framework that takes natural language prompts and produces multimodal training data — scenes, tasks, trajectories, and sensor streams.

Data scarcity is the main bottleneck in robotics. Genesis turns language descriptions into infinite training data.

Photo-realistic ray-tracing rendering

Native ray-tracing-based rendering pipeline for generating visually realistic synthetic images that bridge the sim-to-real gap.

Policies trained on photo-realistic simulation transfer better to real-world deployment.

Install

One command to start

$ pip install genesis-world

Use cases

What teams use it for

Large-scale policy training data generation

Use Genesis to generate millions of demonstration episodes across diverse scenes, tasks, and physical conditions for training robust robot policies.

Sim-to-real transfer research

The combination of physics fidelity, rendering quality, and domain randomization in one framework makes Genesis a strong platform for sim-to-real studies.

Deformable object manipulation

Genesis's MPM, SPH, and FEM solvers enable realistic simulation of deformable objects, fluids, and granular materials for advanced manipulation research.

Ecosystem

Tags & capabilities

botopen sourceroboticsopen source

Comparison

How it stacks up

Choose Genesis for generative data at scale

vs traditional robotics simulators

MuJoCo and PyBullet are excellent for rigid-body simulation but lack Genesis's multi-solver integration, generative AI pipeline, and extreme simulation speed.

FAQ

Questions

What hardware do I need to run Genesis?

Genesis runs on Linux, macOS, and Windows. GPU acceleration (NVIDIA/AMD) provides the best performance, but CPU mode is also available. The 430,000x speed benchmark was on a single RTX 4090.

Does Genesis support my robot?

Genesis supports loading MJCF (.xml), URDF, .obj, .glb, .ply, and .stl files. If your robot has a URDF, it can be simulated in Genesis.

Is Genesis differentiable?

Yes, Genesis is designed to be fully differentiable. Currently the MPM solver and Tool Solver support differentiability, with rigid body differentiation planned.