Open Source RoboticsRobots, software & learning

Genesis World

Software

Genesis World, formerly Genesis, is a Python simulation platform combining multiple physics solvers, robot interfaces and rendering backends; optional features have different platform requirements. [1]

Installation & first result →Official websiteGetting started

For
Robotics developers building simulation and data-generation environments [1]
Source rights
Named licenses are documented by part. Review licenses
Unknowns
compute requirements not fully established. Review gaps

Source researched · Not tested by us
Sources checked · Research details

Licenses & related hardware

Requirements · Research details and dates

On this page

Overview

Genesis World, previously named Genesis, is a Python simulation platform combining rigid and deformable physics, robot interfaces and rendering. It can serve as the environment underneath a robotics learning or data-generation pipeline. This profile reviews release v1.4.4, published October 11, 2026, rather than assuming older Genesis tutorials describe the current stack. [1] [6]

Is this for you?

Known limitations
  • IPC extra is documented for Linux/Windows x86 with an NVIDIA GPU.
  • Renderer and compiler availability does not establish every solver works on every platform.
  • Scene stepping is not a learned-policy benchmark or evidence of physical robot fidelity.
  • Assets have component-specific licenses; submodule payloads were not reviewed.

Editorial assessment from cited documentation; no runtime testing.

[1] [3] [4] [5] [8] · Editorial assessment

What you need

Operating system
  • Linux
  • Windows
  • macOS; feature and backend coverage differs

[1]

Languages / runtimes
  • Python

[2]

Dependencies
  • Python >=3.10,<3.14
  • PyTorch installed separately for the chosen platform
  • Quadrants 1.3.3 in release 1.4.4
  • Optional pyuipc for IPC; gs-nyx for Nyx rendering

[1] [2]

Host computer

CPU execution is shown in the basic scene; accelerated and optional backends need compatible hardware. Complete scene-specific RAM/VRAM and throughput are not established.

[1] [3]

Installation & first result

Open the upstream getting-started guide [3] · Project documentation [1]

  • Install platform-appropriate PyTorch in a supported Python environment.
  • Install the reviewed genesis-world 1.4.4 package or matching source revision.
  • Run the pinned hello_genesis.py CPU example to build a plane and Panda scene and step it 1000 times.
  • Add accelerated or optional rendering/physics backends only after checking their specific requirements.

[1] [2] [3]

Installation

Install PyTorch first, then the release package genesis-world==1.4.4 in Python 3.10–3.13.

Editorial assessment from cited documentation; no runtime testing.

[1] [2] · Editorial assessment

Hardware, models & data

Inputs and outputs

Robot and scene descriptions, materials and control inputs produce simulated state and configured sensor/render outputs.

[1]

File formats
  • URDF
  • MJCF
  • OBJ
  • GLB
  • USD

[1]

Evaluation

Use the CPU scene as an installation check, then separately validate the solver, contacts, controller and rendering used by the target task.

Editorial assessment from cited documentation; no runtime testing.

[3] · Editorial assessment

Project notes

Match the installation to the backend

Release 1.4.4 requires Python 3.10 through 3.13. Install the appropriate PyTorch build first, then the genesis-world package; pin genesis-world==1.4.4 when following this review. Package metadata pins the Quadrants compiler to 1.3.3. A compiler dependency is installed automatically, while some physics and rendering paths require additional packages. [1] [2]

The README lists CPU, CUDA, ROCm, Metal and Vulkan targets through Quadrants, but that broad compiler coverage does not establish that every feature works everywhere. In particular, the optional IPC solver uses pyuipc and is documented for Linux or Windows on x86 with an NVIDIA GPU. The Nyx renderer is a separate gs-nyx installation. Check the selected solver and renderer rather than treating the whole platform as one hardware requirement. [1]

Begin with one CPU scene

The pinned examples/tutorials/hello_genesis.py initializes the CPU backend, adds a plane and the bundled Panda MJCF model, builds the scene and runs 1,000 simulation steps. This is a small first check before parallel environments, complex contacts or rendering experiments. The example does not load a learned policy, measure throughput or validate a physical robot. [3]

The Panda asset documentation describes a simplified model derived from a URDF, with manually edited collision geometry, actuators and constraints. That provenance matters when interpreting simulated contact or controller behavior. A successful scene run would still leave task-specific physical fidelity to establish. We have not run the scene or measured memory, speed or accuracy. [4]

Separate engine, asset and optional-component terms

The engine source is Apache-2.0. The bundled Panda description declares Apache-2.0, while the UR5e asset includes a three-clause BSD-style license. Those examples show why the engine's root license should not be assigned to every robot asset. External components and datasets need their own checks. [7] [4] [5]

The repository also lists documentation, LuisaRender and ParticleMesher as submodules. This review covers the preserved engine source and inspected files; those separate submodule payloads and optional renderer installations were not acquired or tested. Treat the profile as source research, with no simulation benchmark or hardware-transfer claim. [8]

What we checked

Source researched only; no installation, training, benchmark or physical hardware test performed. No affiliate links or reused media.

Not yet established

  • Version-specific hardware compatibility (ecosystem tags alone do not establish it)

Updates

Upstream last updated
Oct 11, 2026 [6]
Latest release
Oct 11, 2026 [6]
Profile updated
Oct 17, 2026
Sources last checked
Oct 17, 2026
Next review
Nov 17, 2026
Reviewed revision
v1.4.4 / a649105285d30bb9cd3464e753bc5270f3f43d4b · software repository

Source files and licenses

Apache-2.0 simulation engine with separate asset and optional-component terms; no blanket license or cross-backend compatibility claim.

LayerAvailability / license
softwareavailable · Apache-2.0
Engine Apache-2.0; bundled assets retain their own notices and licenses. [7] [4] [5]
documentationavailable · Apache-2.0
In-repository documentation only; separate documentation submodule not audited. [7]

The reviewed simulator scene does not require pretrained policy weights.

The reviewed first-scene workflow uses robot assets, not a training dataset.

References

Each source links to the exact revision we read. “Used for” shows where this page relies on it.

Correction guidance · Discussion is not enabled · Profile format 0.7.0-draft.5

Change history