robopal is a multi-platform, modular robot simulation framework based on MuJoCo physics engine, which is mainly used for reinforcement learning training and control algorithm implementation of robotic arms. Please check the Documentation for more information.

robopal 是一个基于 MuJoCo 物理引擎搭建的多平台的，模块化的机器人仿真框架，主要用于机械臂的强化学习训练与控制算法实施。

robopal 为您提供了：

• 采用 Mujoco 原生 API 计算机械臂动力学与运动学，无需额外安装扩展库，提高运行帧数
• 简洁的代码结构，没有复杂的嵌套关系，方便快速上手学习和使用
• 具备 Gymnasium 风格的单臂环境与 PettingZoo 风格的双臂环境，方便集成大部分的单/多智能体强化学习算法库(eg. stable-baselines3，MARL)
• 提供多种基础控制方案，如关节空间/笛卡尔空间的位置控制、速度控制、阻抗控制，并提供了遥操作接口
• 提供丰富的任务环境示例，如 ConveyorBelt，PickAndPlace, Drawer, Cabinet，VisualServo等
• 模块化定制 MJCF 描述的机器人场景模型，可自由组合搭配场景，基座，机械臂，末端执行器和物体

请查看文档以获取更多信息 (更新中)

---

Getting Started

Preparation

• Windows / Linux
• MuJoCo-3.1.4
• Python 3.8 +

Install from pip

You are advised to Install from source to obtain the latest version

$ pip install robopal

Install from source

# Clone robopal
$ git clone https://github.com/NoneJou072/robopal
$ cd robopal
# Install robopal and its requirements.
$ pip install -r requirements.txt

Run a demo

python -m robopal.demos.demo_controller

Contribute

robopal currently has many shortcomings. Everyone is welcome to raise questions or leave suggestions in the Issue, and you are welcome to Pull Request to improve this project together.

Future works

• Documentation and tutorials.
• Teleoperation Interface and rollout function (for Imitation Learning).
• New demos of bimanual, using petting zoo style (for Multi-Agents RL).

Citation

Please cite robopal if you find useful in this work:

@software{Zhou_robopal_A_Simulation_2024,
author = {Zhou, Haoran and Huang, Yichao and Zhao, Yuhan and Lu, Yang},
doi = {10.5281/zenodo.11078757},
month = apr,
title = {{robopal: A Simulation Framework based Mujoco}},
url = {https://github.com/NoneJou072/robopal},
version = {0.3.1},
year = {2024}
}