This repository provides a Simulink® C-MEX S-Function block interface to the MuJoCo™ physics engine.
This blockset enables an easy access to MuJoCo Physics Engine and exposes the GUI, control inputs, sensors and RGB/Depth Cameras from the engine.
- MATLAB® (required)
- Simulink® (required)
- Computer Vision Toolbox™ (optional)
- Robotics System Toolbox™ (optional)
- Control System Toolbox™ (optional)
- Simulink® Coder™ (optional)
MATLAB R2022b or newer is recommended. Install MATLAB with the above products and then proceed to set up MuJoCo blocks. This blockset cannot be installed in MATLAB Online.
Steps for building/rebuilding the C-MEX S-Function code.
-
Install one of the following C++ Compiler
- Microsoft® Visual Studio® 2022 or higher (recommended)
- (or) MinGW (GCC 12.2.0 or higher)
-
Clone this repository or download the source zip (preferably not in Downloads folder)
$ git clone git@github.com:mathworks-robotics/mujoco-simulink-blockset.git -
Launch MATLAB and open the repository folder
>> install
-
Open tools/
- Open setupBuild.m. In case you are using MinGW compiler, edit the file and set selectedCompilerWin to "MINGW".
>> setupBuild>> mex -setup c++>> build
-
Install the tools required for compiling the S-Function
$ sudo apt update && sudo apt install build-essential git libglfw3 libglfw3-dev -
Clone this repository or download the source zip (preferably not in Downloads folder)
$ git clone git@github.com:mathworks-robotics/mujoco-simulink-blockset.git -
Launch MATLAB and open the repository folder. Run the install.m script.
>> install
-
The default pathdef.m is likely not saveable in Linux. Save pathdef.m to new location as given in this MATLAB answer.
savepath ~/Documents/MATLAB/pathdef.m -
Open tools/ and run the following commands in MATLAB command Windows
>> setupBuild>> mex -setup c++>> build
>>mj_gettingStarted
Open the example model and run it in normal simulation mode.
If the installation is successful, you should see a pendulum model running in a separate window and camera streams displayed by Video Viewer blocks (Computer Vision Toolbox).
A dedicated graphics card is recommended for the best performance. Disable Video Viewer blocks if the model runs slow
(Linux users) - In case MATLAB crashes, it may be due to a glibc bug. Please follow this bug report for a workaround!
MuJoCo Plant block steps MuJoCo engine, renders visualization window & camera, sets actuator data, and outputs sensor readings
It takes an XML (MJCF) as the main block parameter. It auto-detects the inputs, sensors, and cameras from XML to configure the block ports and sample time.
Inputs can either be a Simulink Bus or a vector.
Sensors are output as a Simulink Bus.
RGB and Depth buffers from cameras are output as vectors. These can be decoded to Simulink image/matrix using the RGB and Depth Parser blocks.
getting_started_clip1.mp4
getting_started_clip2.mp4
Refer to https://github.com/vmanoj1996/mujoco-demo for a more complex example.
- MuJoCo version update - Open install.m and change MJ_VER to the desired MuJoCo physics engine version.
- Code generation - The MuJoCo Plant block supports code generation (Simulink Coder) and monitor and tune for host target. Refer to mj_monitorTune.slx for more info.
- Performance improvement - In case you want to reduce the mask initialization overhead, you can directly use the underlying S-Function. Select the MuJoCo Plant block and Ctrl+U to look under the subsystem mask. Make sure to call the initialization functions (whenever the MJCF XML model changes).
Linux Compatibility:
This blockset is only tested in Ubuntu 22.04 and Ubuntu 20.04. Other Ubuntu versions and distros are not supported.
This blockset does not work with software OpenGL. You can check whether MATLAB is using hardware GL with >>opengl info command.
In case you face graphics related issues, please try updating GLFW following the instructions given below!
In case MATLAB crashes while running getting started model and you see the following lines in stack trace,
#10 0x00007fdaf8619f40 in glfwCreateWindow () at /lib/x86_64-linux-gnu/libglfw.so.3
#11 0x00007fdaf8675c4d in MujocoGUI::initInThread(offscreenSize*, bool),
Updating glfw could fix the issue.
Building glfw from source (glfw main - commit id) and installing fixed this issue for me,
- sudo apt remove libglfw3 libglfw3-dev # Remove existing glfw
- mkdir ~/glfwupdated
- cd ~/gflwupdated
- git clone git@github.com:glfw/glfw.git
- sudo apt install cmake-qt-gui
- cmake-gui
- Set the source directory to the root of cloned repo
- mkdir build in the cloned repo
- Set the build directory
- In Cmake Gui settings - Select "BUILD_SHARED_LIBS" as well
- Configure and then generate
- Open a terminal in the build directory and run $make in terminal
- Once build goes through without any error, run $sudo make install in terminal
- sudo ldconfig (to refresh linker cache)
- Follow the build instructions for mujoco-simulink-blockset
The license is available in the license.txt file within this repository.
Cite this work as,
Manoj Velmurugan. Simulink Blockset for MuJoCo Simulator (https://github.com/mathworks-robotics/mujoco-simulink-blockset), GitHub. Retrieved date.
Refer to the MuJoCo repository for guidelines on citing the MuJoCo physics engine.
The sample codes and API documentation provided for MuJoCo and GLFW were used as reference material during development.
MuJoCo and GLFW libraries are dynamically linked against the S-Function and are required for running this blockset.
UR5e MJCF XML from MuJoCo Menagerie was used for creating demo videos.
You can post your queries in the discussions section.
Copyright 2023 The MathWorks, Inc.
