GGLBot

Creates everything you need to get your bot made with GigaLearn (GGL) up and running in RLBot v5.

Features

Creates a stripped down .exe file (~3MB) that runs your GGL bot without all the learning/optimization fluff. All you need to do is add your Obs Builder, Action Parser, InferUnit config, and .lt models. Additionally, since the agent_id in the .exe is set by the bot.toml file and libtorch is in a central location, after you build your .exe, you can copy/paste the entire rlbot\ folder after building. Then you only need to change the bot.toml and model files to quickly add different versions of your bot to RLBot - as long as they use the same obs/parser/model setup.

Instructions

• Clone this repo recursively: git clone https://github.com/SubparN0va/GGLBot --recurse-submodules
• Update RLBotMain.cpp with your Obs Builder, Action Parser, and InferUnit config
  • If creating new obs or parser files, make sure you update the #include at the top of RLBotClient.h
• Make sure the project can locate libtorch (see note below)
• Build using Release mode. This will automatically place the .exe into the rlbot\ folder
• Put your models (.lt files) into the rlbot\ folder
• Update the bot.toml and loadout.toml to your preference
• In RLBot v5, add the rlbot\ folder

Note

This project expects the cpu version of libtorch to be located in %LOCALAPPDATA%\RLBot5\bots\libtorch_cpu. Eventually, RLBot v5 may ship with this, but for now you can add it manually. If you want to put it in a different location, make sure you update both the rlbot\run.bat (bot.toml uses run.bat to set the path to libtorch before opening the .exe) and CMakeLists.txt to point to it. If submitting a bot to a tournament, leave this at the default location.