Recurrent PPO is fully in C++ using libtorch to master several sub levels of Doom by just looking at the pixel information.

During the last decade I couldn't find efficient implementation of Prioritized Experience Replay (PER). I decided to initiate my own. Now it is a python package for RL engineering.

Very often people desire to play games that have multiple mixed contents from their past experience but there is no good way of searching for it. Now you can simply search your game with a simple sentence like "tank simulation game in space with unicorns".