Skip to content

Genesis World Setup / Genesis World 环境

Environment hub · 中文指南

English guide

The maintained project and PyPI distribution are named Genesis World and genesis-world. Follow the current installation guide, not a command copied from the earlier repository naming.

Install

  1. Create a clean Python environment.
  2. Install PyTorch with the official platform selector.
  3. Install Genesis World:
python -m pip install genesis-world

Check the current Python and platform range on the official page before creating the environment. Rendering extras and source-development extras are separate choices; install them only when the selected workflow requires them.

Minimal CPU smoke

import genesis as gs

gs.init(backend=gs.cpu, seed=0)
scene = gs.Scene(show_viewer=False)
scene.add_entity(gs.morphs.Plane())
scene.build()
for _ in range(10):
    scene.step()
print("Genesis scene stepped")

This follows the official sequence: initialize once, create a scene, add entities, build, and step. First execution can include compilation work; record warm-up separately from steady-state timing.

GPU and parallel environments

Verify three facts independently:

  • the selected PyTorch build can see the intended accelerator;
  • Genesis reports the selected backend rather than silently falling back;
  • the scene and tensor batch dimensions match the number of environments.

Do not convert a vendor speed headline into a local throughput claim. Report scene complexity, backend, warm-up, number of environments, control decimation, rendering state, and measured steps per second on the local machine.

中文指南

当前维护的项目与 PyPI 包名是 Genesis World 和 genesis-world。先根据 PyTorch 官方选择器安装与平台匹配的 PyTorch，再执行 python -m pip install genesis-world；Python 与平台范围以当前官方安装页为准。

最小程序遵循 gs.init → Scene → add_entity → build → step。第一次执行可能包含编译开销，因此预热时间与稳态吞吐必须分开记录。GPU 路径还要分别检查 PyTorch 是否看到加速器、Genesis 是否选择了预期后端、并行环境批维是否正确。官方宣传数字不能替代本机测量；报告吞吐时需同时记录场景复杂度、后端、环境数、控制降频和渲染状态。

权威来源：Installation · Initialization · Hello Genesis