为具身智能和智能体而生的强化学习框架

开源的具身智能与智能体强化学习框架。PPO、GRPO、SAC 等算法开箱即用，VLA 模型从仿真训练到真机部署一条链路打通。

快速开始

运行在业界领先厂商的硬件上：

具身与智能体，一套框架
VLA 和世界动作模型可以在仿真或真机上训练，智能体支持单智能体和多智能体强化学习，都在同一套框架里完成。
扩展不用改代码
分布式的复杂度交给框架：扩到大量 GPU 节点、在 FSDP 和 Megatron 后端之间切换，都不用改代码。
国产和国外硬件都能跑
同一套训练栈可以运行在 NVIDIA、AMD GPU 以及华为昇腾、摩尔线程、沐曦、壁仞等国产加速卡上，迁移时模型和算法零改动。

几秒钟上手

用 UV 或 Docker 安装 RLinf，再按教程跑通第一次训练。

安装 RLinf
在右边选好配置，运行对应的命令。
跑通 VLA 快速上手
按快速上手教程训练第一个 VLA 模型。
扩展到多机
把同样的训练扩展到多台机器上运行。

安装文档

安装方式

场景

模型

硬件

下载源

安装命令

git clone https://github.com/RLinf/RLinf.git
cd RLinf
bash requirements/install.sh embodied --model openvla-oft --env maniskill_libero
source .venv/bin/activate

git clone https://github.com/RLinf/RLinf.git
cd RLinf
bash requirements/install.sh embodied --model openvla-oft --env maniskill_libero --use-mirror
source .venv/bin/activate

git clone https://github.com/RLinf/RLinf.git
cd RLinf
bash requirements/install.sh --platform amd embodied --model openvla-oft --env maniskill_libero
source .venv/bin/activate

git clone https://github.com/RLinf/RLinf.git
cd RLinf
bash requirements/install.sh --platform amd embodied --model openvla-oft --env maniskill_libero --use-mirror
source .venv/bin/activate

git clone https://github.com/RLinf/RLinf.git
cd RLinf
bash requirements/install.sh --platform ascend embodied --model openvla-oft --env maniskill_libero
source .venv/bin/activate

git clone https://github.com/RLinf/RLinf.git
cd RLinf
bash requirements/install.sh --platform ascend embodied --model openvla-oft --env maniskill_libero --use-mirror
source .venv/bin/activate

git clone https://github.com/RLinf/RLinf.git
cd RLinf
bash requirements/install.sh --platform musa embodied --model openvla-oft --env libero
source .venv/bin/activate

git clone https://github.com/RLinf/RLinf.git
cd RLinf
bash requirements/install.sh --platform musa embodied --model openvla-oft --env libero --use-mirror
source .venv/bin/activate

git clone https://github.com/RLinf/RLinf.git
cd RLinf
bash requirements/install.sh embodied --model openvla --env maniskill_libero
source .venv/bin/activate

git clone https://github.com/RLinf/RLinf.git
cd RLinf
bash requirements/install.sh embodied --model openvla --env maniskill_libero --use-mirror
source .venv/bin/activate

git clone https://github.com/RLinf/RLinf.git
cd RLinf
bash requirements/install.sh --platform amd embodied --model openvla --env maniskill_libero
source .venv/bin/activate

git clone https://github.com/RLinf/RLinf.git
cd RLinf
bash requirements/install.sh --platform amd embodied --model openvla --env maniskill_libero --use-mirror
source .venv/bin/activate

git clone https://github.com/RLinf/RLinf.git
cd RLinf
bash requirements/install.sh --platform ascend embodied --model openvla --env maniskill_libero
source .venv/bin/activate

git clone https://github.com/RLinf/RLinf.git
cd RLinf
bash requirements/install.sh --platform ascend embodied --model openvla --env maniskill_libero --use-mirror
source .venv/bin/activate

git clone https://github.com/RLinf/RLinf.git
cd RLinf
bash requirements/install.sh --platform musa embodied --model openvla --env libero
source .venv/bin/activate

git clone https://github.com/RLinf/RLinf.git
cd RLinf
bash requirements/install.sh --platform musa embodied --model openvla --env libero --use-mirror
source .venv/bin/activate

git clone https://github.com/RLinf/RLinf.git
cd RLinf
bash requirements/install.sh embodied --model openpi --env maniskill_libero
source .venv/bin/activate

git clone https://github.com/RLinf/RLinf.git
cd RLinf
bash requirements/install.sh embodied --model openpi --env maniskill_libero --use-mirror
source .venv/bin/activate

git clone https://github.com/RLinf/RLinf.git
cd RLinf
bash requirements/install.sh --platform amd embodied --model openpi --env maniskill_libero
source .venv/bin/activate

git clone https://github.com/RLinf/RLinf.git
cd RLinf
bash requirements/install.sh --platform amd embodied --model openpi --env maniskill_libero --use-mirror
source .venv/bin/activate

git clone https://github.com/RLinf/RLinf.git
cd RLinf
bash requirements/install.sh --platform ascend embodied --model openpi --env maniskill_libero
source .venv/bin/activate

git clone https://github.com/RLinf/RLinf.git
cd RLinf
bash requirements/install.sh --platform ascend embodied --model openpi --env maniskill_libero --use-mirror
source .venv/bin/activate

git clone https://github.com/RLinf/RLinf.git
cd RLinf
bash requirements/install.sh --platform musa embodied --model openpi --env libero
source .venv/bin/activate

git clone https://github.com/RLinf/RLinf.git
cd RLinf
bash requirements/install.sh --platform musa embodied --model openpi --env libero --use-mirror
source .venv/bin/activate

git clone https://github.com/RLinf/RLinf.git
cd RLinf
bash requirements/install.sh embodied --model gr00t --env maniskill_libero
source .venv/bin/activate

git clone https://github.com/RLinf/RLinf.git
cd RLinf
bash requirements/install.sh embodied --model gr00t --env maniskill_libero --use-mirror
source .venv/bin/activate

git clone https://github.com/RLinf/RLinf.git
cd RLinf
bash requirements/install.sh --platform amd embodied --model gr00t --env maniskill_libero
source .venv/bin/activate

git clone https://github.com/RLinf/RLinf.git
cd RLinf
bash requirements/install.sh --platform amd embodied --model gr00t --env maniskill_libero --use-mirror
source .venv/bin/activate

git clone https://github.com/RLinf/RLinf.git
cd RLinf
bash requirements/install.sh --platform ascend embodied --model gr00t --env maniskill_libero
source .venv/bin/activate

git clone https://github.com/RLinf/RLinf.git
cd RLinf
bash requirements/install.sh --platform ascend embodied --model gr00t --env maniskill_libero --use-mirror
source .venv/bin/activate

git clone https://github.com/RLinf/RLinf.git
cd RLinf
bash requirements/install.sh --platform musa embodied --model gr00t --env libero
source .venv/bin/activate

git clone https://github.com/RLinf/RLinf.git
cd RLinf
bash requirements/install.sh --platform musa embodied --model gr00t --env libero --use-mirror
source .venv/bin/activate

git clone https://github.com/RLinf/RLinf.git
cd RLinf
bash requirements/install.sh agentic --engine sglang
source .venv/bin/activate

git clone https://github.com/RLinf/RLinf.git
cd RLinf
bash requirements/install.sh agentic --engine sglang --use-mirror
source .venv/bin/activate

git clone https://github.com/RLinf/RLinf.git
cd RLinf
bash requirements/install.sh agentic --engine vllm
source .venv/bin/activate

git clone https://github.com/RLinf/RLinf.git
cd RLinf
bash requirements/install.sh agentic --engine vllm --use-mirror
source .venv/bin/activate

docker pull rlinf/rlinf:agentic-rlinf0.4-maniskill_libero
docker run -it --gpus all \
  --shm-size 100g \
  --net=host \
  --name rlinf \
  -e NVIDIA_DRIVER_CAPABILITIES=all \
  rlinf/rlinf:agentic-rlinf0.4-maniskill_libero /bin/bash

git clone https://github.com/RLinf/RLinf.git
cd RLinf
source switch_env openvla-oft

docker pull rlinf/rlinf:agentic-rlinf0.4-maniskill_libero
docker run -it --gpus all \
  --shm-size 100g \
  --net=host \
  --name rlinf \
  -e NVIDIA_DRIVER_CAPABILITIES=all \
  rlinf/rlinf:agentic-rlinf0.4-maniskill_libero /bin/bash

git clone https://github.com/RLinf/RLinf.git
cd RLinf
source switch_env openvla

docker pull rlinf/rlinf:agentic-rlinf0.4-maniskill_libero
docker run -it --gpus all \
  --shm-size 100g \
  --net=host \
  --name rlinf \
  -e NVIDIA_DRIVER_CAPABILITIES=all \
  rlinf/rlinf:agentic-rlinf0.4-maniskill_libero /bin/bash

git clone https://github.com/RLinf/RLinf.git
cd RLinf
source switch_env openpi

docker pull rlinf/rlinf:agentic-rlinf0.4-maniskill_libero
docker run -it --gpus all \
  --shm-size 100g \
  --net=host \
  --name rlinf \
  -e NVIDIA_DRIVER_CAPABILITIES=all \
  rlinf/rlinf:agentic-rlinf0.4-maniskill_libero /bin/bash

git clone https://github.com/RLinf/RLinf.git
cd RLinf
source switch_env gr00t

架构

广泛的算法、模型与硬件支持

RLinf 支持的强化学习算法、模型、仿真器和硬件。

硬件

NVIDIA GPU
AMD GPU
华为昇腾
摩尔线程
沐曦
壁仞

查看全部硬件（在新标签页打开）

版本发布

v0.3

最新218 个 PR · 65 位贡献者 · 34 位新贡献者

2026年7月15日

新模型

Dexbotic DM0
DreamZero
GR00T-N1.6 / N1.7
ABot-M0
StarVLA
LingBot-VLA

亮点

新增 6 个具身模型，包括基于世界模型的 VLA DreamZero，以及 GR00T-N1.6 / N1.7、StarVLA、LingBot-VLA。文档（在新标签页打开）
新增 Genesis、Polaris、RoboVerse 仿真器，以及 LIBERO+ / LIBERO-Pro、Embodichain、RoboCasa 示例。文档（在新标签页打开）
真机全链路打通：数据采集 → SFT → RL → 部署，支持 Spacemouse、VR、GELLO 三种遥操作采集。文档（在新标签页打开）
新增真机和仿真强化学习算法，包括面向 π0.5 的 DSRL、RECAP、SAC-Flow、DAgger 和 HG-DAgger。文档（在新标签页打开）
通过 install.sh --platform 在华为昇腾（CANN）和 AMD ROCm 上端到端运行。文档（在新标签页打开）
新增 SGLang 推理服务、环境解耦模式，并升级了权重同步。文档（在新标签页打开）

还有 2 条亮点

在 GitHub 查看全部版本（在新标签页打开）

加入社区

从第一次上手的新用户，到做大规模训练的团队，社区向所有人开放。

为具身智能和智能体而生的强化学习框架

具身与智能体，一套框架

扩展不用改代码

国产和国外硬件都能跑

几秒钟上手

安装 RLinf

跑通 VLA 快速上手

扩展到多机

架构

广泛的算法、模型与硬件支持

强化学习算法

模型

仿真器

硬件

版本发布

v0.3

新模型

亮点

v0.2

亮点

v0.1

亮点

加入社区

为具身智能和智能体而生的强化学习框架

特性

具身与智能体，一套框架

扩展不用改代码

国产和国外硬件都能跑

几秒钟上手

安装 RLinf

跑通 VLA 快速上手

扩展到多机

架构

广泛的算法、模型与硬件支持

强化学习算法

模型

仿真器

硬件

版本发布

新模型

亮点

亮点

亮点

加入社区