Cartesian Controller Simulation

This package provides a simulated robot for controller development and testing.

Physics engine and rationales

We build the simulator on Todorov’s MuJoCo physics engine, which has been acquired and open-sourced by Google here. This gives us a strong environment to realistically test control and contact phenomena with minimal dependencies.

Build and install

We use MuJoCo in headless mode and don’t need OpenGL-related dependencies.

1. Download MuJoCo’s pre-built library package and extract that somewhere. It’s ready-to-use and we will just point to it during the build.

   cd $HOME
   wget https://github.com/deepmind/mujoco/releases/download/3.0.0/mujoco-3.0.0-linux-x86_64.tar.gz
   tar -xf mujoco-3.0.0-linux-x86_64.tar.gz
   

1. Switch to the root of your ROS2 workspace and build the package (standalone) with

   colcon build --cmake-args "-DMUJOCO_DIR=$HOME/mujoco-3.0.0" --packages-select cartesian_controller_simulation
   

Getting started

In a sourced terminal, run

export LC_NUMERIC="en_US.UTF-8"
ros2 launch cartesian_controller_simulation simulation.launch.py

The export might not be necessary on your system. It fixes an eventual locals problem and makes sure that you see the robot visualization correctly in RViz. The launch part will start a simulated world with a generic robot model. You can call

ros2 control list_controllers

to get a list of controllers currently managed by the controller_manager. All of them can be activated and tested in the simulator. In contrast to ROS1, these controllers are nodes and you can also see them with ros2 node list.

Hardware interfaces for controllers

Exposed interfaces per joint:

• command_interfaces: position, velocity, stiffness, damping
• state_interfaces: position, velocity