本站点为独立社区,非 Hugging Face 官方。官方仓库:https://github.com/huggingface/lerobotThis site is an independent community hub, not affiliated with Hugging Face. Official repo: https://github.com/huggingface/lerobot
LeRobot 如何在 15 分钟内完成安装并验证环境?How to Install and Verify LeRobot in 15 Minutes?
立即答案:用 Conda 固定 Python 版本,按官方仓库安装依赖,并用最小验证脚本输出版本号即可确认安装成功。Direct Answer: Pin Python version with Conda, install dependencies from official repo, and verify by outputting the version number.
TL;DR / Key Takeaways
- 1.必须使用 Python 3.10+ 环境以确保兼容性。Must use Python 3.10+ environment for compatibility.
- 2.推荐在 Linux (Ubuntu 22.04) 上运行,macOS 仅支持部分仿真。Recommended on Linux (Ubuntu 22.04), macOS supports partial simulation only.
- 3.硬件连接前需安装 `feetech-servo-sdk` (针对 SO-100)。Install `feetech-servo-sdk` before connecting hardware (for SO-100).
LeRobot 支持哪些硬件?我应该先选 SO-100 还是 Aloha/Unitree?Supported Hardware: SO-100, Aloha, or Unitree?
立即答案:想最快跑通闭环就选 SO-100;要做更复杂动作与更高可靠性,再升级到 Aloha 或 Unitree。Direct Answer: Choose SO-100 for the fastest end-to-end setup. Upgrade to Aloha/Unitree for complex tasks.
| 硬件名称Hardware | 适用场景Use Case | 成本 (预估)Est. Cost | 难度Difficulty |
|---|---|---|---|
| SO-100 | 单臂模仿学习、初学者Single-arm IL, Beginners | $200 - $300 | 入门级Entry Level |
| Aloha | 双臂协同、精细操作Dual-arm, Fine Manipulation | $10k+ | 进阶级Advanced Level |
| Unitree G1 | 人形机器人、强化学习Humanoid, RL | $16k+ | 专家级Expert Level |
数据来源:Hugging Face LeRobot 官方仓库 + 硬件厂商官网 | 更新于:2026-03-18
我该用 Linux 还是 macOS/Windows 来安装 LeRobot?Should I use Linux, macOS, or Windows to install LeRobot?
立即答案:真实硬件优先 Linux;macOS/Windows 更适合先跑仿真与阅读教程。Direct Answer: Prefer Linux for real hardware. macOS/Windows are better suited for running simulations and reading tutorials first.
| 系统 | 真实硬件支持 | 推荐用途 | 风险点 |
|---|---|---|---|
| Ubuntu 22.04 | 强 | 真机训练/部署 | 驱动/权限 |
| macOS | 有限 | 仿真/数据处理 | 硬件驱动缺失 |
| Windows | 有限 | 先学流程 | WSL/驱动差异 |
数据来源:社区实测 | 更新于:2026-03-18
安装前我必须检查哪些环境依赖,才能避免 80% 报错?What environment dependencies must I check before installation to avoid 80% of errors?
立即答案:固定 Python 版本、确认 GPU/驱动、预留磁盘空间并确保编译工具链可用。Direct Answer: Pin the Python version, confirm GPU/drivers, reserve disk space, and ensure the build toolchain is available.
软件要求
- • Ubuntu 20.04/22.04 (推荐)
- • Python 3.10
- • Conda / Mamba
- • Git
硬件要求
- • NVIDIA GPU (建议 8GB+ 显存)
- • 16GB RAM
- • 50GB 可用磁盘空间
没有 GPU 也能跑吗?本地 vs 云 GPU 我该怎么选?Can I run it without a GPU? Local vs. Cloud GPU: How to choose?
立即答案:能跑。小规模验证用 CPU/低端 GPU;要稳定训练与提速,建议云 GPU。Direct Answer: Yes. Use CPU/low-end GPU for small-scale validation; for stable training and speedup, cloud GPUs are recommended.
| 方案 | 速度 | 成本 | 适用阶段 |
|---|---|---|---|
| CPU | 慢 | 最低 | 流程验证/调参 |
| 本地 GPU(8GB+) | 中 | 一次性硬件 | 稳定训练 |
| 云 GPU | 快 | 按需付费 | 冲刺阶段 |
数据来源:社区实测 | 更新于:2026-03-18
LeRobot 安装步骤怎么做才算“可复现”?How to make LeRobot installation steps "reproducible"?
立即答案:每一步都要有可验证输出(版本号/命令返回/设备枚举),不要跳步。Direct Answer: Every step must have a verifiable output (version number/command return/device enumeration). Do not skip steps.
步骤 1: 克隆仓库与创建环境
git clone https://github.com/huggingface/lerobot.git cd lerobot conda create -y -n lerobot python=3.10 conda activate lerobot
步骤 2: 安装依赖包
pip install -e . pip install opencv-python # 可选,用于相机支持
步骤 3: 验证安装
python -c "import lerobot; print(lerobot.__version__)"
提示:看到版本号输出即表示安装成功!
常见安装错误及解决方案Common Installation Errors & Solutions
错误:Error: 'ModuleNotFoundError: No module named 'feetech_servo_sdk''
解决: 请确保运行了 `pip install -e .`。如果仍报错,请手动安装 sdk 子模块。Solution: Make sure you ran `pip install -e .`. If it still fails, install the sdk submodule manually.
错误:Error: `CUDA Error: no kernel image is available`
解决: 你的显卡驱动版本过低或 PyTorch 版本不匹配。请检查 `nvidia-smi` 输出。Solution: Your GPU driver is too old or PyTorch version mismatches. Check `nvidia-smi` output.