# 智元 Genie Sim 安装、场景与基准任务运行实操

> 原文链接：https://robotics.hellobike.com/tutorials/agibot-genie-sim-quickstart
> 频道：tutorials | 更新日期：2026-10-22 | 标签：Genie Sim, geniesim_cli, Benchmark, RT Engine, ROS 2, 场景生成
> 核心摘要：本文按 Genie Sim README 梳理从 CLI 安装、Docker 容器启动、健康检查，到 Benchmark、RT Engine、3D world、场景重建与提示词生成场景的操作路径，便于直接照着执行。

# 智元 Genie Sim 安装、场景与基准任务运行实操

> 本文依据以下官方文档整理：[raw.githubusercontent.com/AgibotTech/genie_sim/main/README.m](https://raw.githubusercontent.com/AgibotTech/genie_sim/main/README.md)。命令与版本以文档当前版本为准。

## 前置条件

- 在 host 上使用 pip 安装 `source/geniesim_cli/`
- 使用 `geniesim docker build` 构建 Isaac Sim 5.1 + Jazzy image
- Benchmark 需要有自己的 inference server 用于本地调试
- `geniesim_assets` 通过 HuggingFace / ModelScope out of band 分发
- tier-2 peers `geniesim_teleop`、`geniesim_generator`、`geniesim_world` 为可选安装，涉及 VR / LLM / CUDA-ML heavy stacks

## 1. 安装 Genie Sim CLI 并构建容器

先在 host 侧安装独立 CLI wheel，然后构建 Isaac Sim 5.1 + Jazzy image。README 将 `geniesim` 作为整个平台入口，用于 docker lifecycle、ROS 2 workspace builds、doctor、bootstrap 等。

```bash
pip install -e source/geniesim_cli/
geniesim docker build
```

## 2. 启动容器并进入交互 shell

容器启动时，entrypoint 会从 bind-mounted workspace editable-installs tier-1 peers。随后进入容器交互 shell，后续 RT Engine、Benchmark、场景相关命令都按文档在容器内继续操作。

```bash
geniesim docker up
geniesim docker into
```

## 3. 在容器内做健康检查与 bootstrap

进入容器后先执行状态探测和诊断修复。`geniesim status` 是 per-distribution health probe，`geniesim doctor` 用于 diagnose & repair，`geniesim bootstrap` 可重装 tier-1 peers 并选择 tier-2。

```bash
geniesim status
geniesim doctor
geniesim bootstrap
```

## 4. 运行 Genie Sim Benchmark 前查看用法

Genie Sim Benchmark 用于本地调试并可衔接 RoboColiseum。文档要求针对自己的 inference server 在本机运行 benchmark task；具体 CLI verbs、common commands、overrides 先查看 USAGE。

```bash
cat source/geniesim_benchmark/USAGE.md
```

## 5. 构建 RT Engine 并查看场景启动配方

RT Engine 通过 `geniesim_ros` 将 physics、render、robot 接入 ROS 2。先构建 ROS workspace，再查看 launch scene、MoveIt 2 + WBC RViz、add robot 等 SKILL 配方。

```bash
geniesim ros build dev
cat source/geniesim_ros/skills/build-workspace/SKILL.md
cat source/geniesim_ros/skills/launch-scene/SKILL.md
cat source/geniesim_ros/skills/moveit-wbc/SKILL.md
cat source/geniesim_ros/skills/add-robot/SKILL.md
cat source/geniesim_ros/skills/*/SKILL.md
```

## 6. 查看世界生成、场景重建与提示词生成场景

`geniesim_world` 可由 panorama 生成 PLY / Gaussians；`scene_reconstruction` 处理 LiDAR 与 images；`geniesim_generator` 将自然语言 prompt 编译为 `scene.usda` 与 layout graph。

```bash
cat source/geniesim_world/skills/generate-world/SKILL.md
cat source/scene_reconstruction/AGENTS.md
cat source/geniesim_generator/skills/deploy-generator/SKILL.md
cat source/geniesim_generator/skills/search-assets/SKILL.md
cat source/geniesim_generator/skills/generate-scene/SKILL.md
```

## 验证是否跑通

安装与容器链路可通过 `geniesim status`、`geniesim doctor` 确认；RT Engine 可通过 `geniesim ros build dev` 完成构建并继续按 `source/geniesim_ros/skills/launch-scene/SKILL.md` 启动场景；Benchmark 先按 `source/geniesim_benchmark/USAGE.md` 在本机 against your inference server 运行 benchmark task。

## 文档里点名的注意事项

- Tier-1 peers come up automatically; tier-2 peers are opt-in.
- `geniesim_assets` is a tier-1 peer distributed out of band，sourced from HuggingFace / ModelScope，并由 entrypoint editable-installed。
- Tier-2 peers (`geniesim_teleop` · `geniesim_generator` · `geniesim_world`) have heavy stacks (VR / LLM / CUDA-ML) and are not auto-installed.
- Install with `pip install -e "source/geniesim/[teleop|generator|world|all]"` — or run `geniesim bootstrap` for an interactive re-install of the whole tree.
- Loading the output into the RT Engine is a W.I.P.; generated PLY / Gaussians are consumed by `geniesim_world`'s own tooling.
- Generated scenes are written under `source/geniesim_generator/src/benchmark/config/llm_task/<scene_id>/<run_id>/`; Docker stack 内同样出现在 `/opt/geniesim_generator/src/benchmark/config/llm_task/<scene_id>/<run_id>/`。
- Open the resulting `scene.usda` with Isaac Sim inside the Genie Sim container so its payload paths resolve as generated; if opened from host side, adjust payload paths inside the USDA first.

## 相关 SDK 条目

- [genie_sim](/sdks/agibottech-genie-sim)

## 原始参考来源 (Sources)

- [raw.githubusercontent.com/AgibotTech/genie_sim/main/README.md](https://raw.githubusercontent.com/AgibotTech/genie_sim/main/README.md)