The Maximum A Psoteriori Policy Optimization algorithm written in C++ with LibTorch.
Mostly translated from the implentation in Python by acyclics. Here is the Papers With Code link.
I wrote this in C++ because MuJoCo (the simulator) is in C++ and robotic componant hardware drivers are in C++. Why slow down in Python when we can go 100% C++? I found any mujoco models with complexity with many DOF, like the humanoid impossible to train. Maybe it's because I just have one 3090. Maybe it's because there is a problem with the code. I welcome any feedback in the 'issues'.
To install LibTorch git clone the PyTorch repo, find the LibTorch directory and build from source.
Other dependancies are
mujoco
numcpp
gsl
fmt
glfw
To install these build from source and make sure to run cmake --install . after building to install on your system at /usr/local if using Linux.
I've not tried this in Windows in Visual Studio. I'm sure it's possible.