For safe and reliable deployment of any robot controller on the real hardware platform, it is generally a necessary practice to comprehensively assess the performance of the controller with the specific robot in a realistic simulation environment beforehand. While there exist several software solutions that can provide the core physics engine for this purpose, it is often a cumbersome and error-prone effort to interface the simulation environment with the robot controller being evaluated. The controller may have a complex structure consisting of multiple states and transitions within a finite-state machine (FSM), and may even require input through a GUI. In this work, we present mc-mujoco -- an open-source software framework that forms an interface between the MuJoCo physics simulator and the mc-rtc robot control framework. We provide implementation details and describe the process for adding support for essentially any new robot. We also demonstrate and publish a sample FSM controller for bipedal locomotion and stable grasping of a rigid object by the HRP-5P humanoid robot in MuJoCo. The code and usage instructions for mc-mujoco, the developed robot modules, and the FSM controller are available online.
翻译:对于在实际硬件平台上安全可靠地部署任何机器人控制器来说,一般来说,在现实的模拟环境中,全面评估控制器与特定机器人在现实模拟环境中的性能是必要的做法。虽然存在若干软件解决方案,可以为此提供核心物理引擎,但将模拟环境与正在评价的机器人控制器连接起来往往是一种繁琐和容易出错的努力。控制器可能有一个复杂的结构,由多状态和在有限状态机器(FSM)中的过渡构成,甚至需要通过一个图形界面输入。在这项工作中,我们介绍了Mc-mujoco -- -- 一个在 Mujoco物理模拟器和 mc-rock机器人控制框架之间形成界面的开放源软件框架。我们提供了实施细节,并描述了基本上为任何新机器人增加支持的程序。我们还演示和出版一个FSM控制器样本,用于在MuJooco中由HRP-5P类人类机器人固定地捕捉硬性物体。Mc-mujoco、开发的机器人模块和FSMSM控制器的代码和使用指令是在线提供的。