GigaLearn for RLBot Integration

This repository provides the necessary files to run a GigaLearnCPP agent within the RLBot framework for Rocket League.

The main purpose is to bridge the GigaLearn inference model with the RLBot client, allowing your trained bot to play in standard RLBot matches and events.

Prerequisites

Before you begin, you should have:

• A working, compiled GigaLearnCPP project.
• RLBot installed and running on your machine.

Setup Instructions

Step 1: Replace Core Project Files

You will need to replace files in your GigaLearnCPP directory to integrate the RLBot logic.

• Copy CMakeLists.txt:

  • Source: CMakeLists.txt from this repository.
  • Destination: replace GigaLearnCPP\CMakeLists.txt.

• Copy Source Files:

  • Source: RLBotClient.h, RLBotClient.cpp, and rlbotmain.cpp from this repository.
  • Destination: Place these in GigaLearnCPP\src\, replacing any existing files.

Step 2: Configure the RLBot Agent

• Add the Agent Config:

  • Source: CppPythonAgent.cfg from this repository.
  • Destination: GigaLearnCPP\rlbot\CppPythonAgent.cfg

• Set Your Executable Path: You must edit CppPythonAgent.cfg to point to your compiled bot's executable.

  1. Open GigaLearnCPP\rlbot\CppPythonAgent.cfg.
  2. Find the line cpp_executable_path = .
  3. Add the full path to your compiled rlbot.exe file.

  Example:

  [Locations]
  # Path to the C++ executable
  cpp_executable_path = C:\path\to\your\GigaLearnCPP\build\Release\rlbot.exe

How to Run

1. After replacing the files, re-compile your GigaLearnCPP project and use regular build tool (ex: Visual Studio).
2. Ensure you have a checkpoints folder with your trained model inside the same directory as your final executable (e.g., GigaLearnCPP\build\Release\). The bot will automatically load the latest model.
3. Add the GigaLearnCPP\rlbot\CppPythonAgent.cfg to rlbot.
4. Launch a match from the RLBot GUI, and your bot should appear.

Optional: Using a Specific Checkpoint

By default, the bot loads the latest checkpoint. To force it to use a specific model, you can hardcode the path.

1. Open GigaLearnCPP\src\rlbotmain.cpp.
2. Find the commented-out line that begins with // checkpointPath =.
3. Uncomment it and replace the path with the full path to your POLICY.lt file.

Example:

// On Windows, you must use double backslashes (\\) or forward slashes (/) in the path.
checkpointPath = "C:\\Users\\YourUser\\GigaLearnCPP\\build\\Release\\checkpoints\\14594451456\\POLICY.lt";

Troubleshooting

• Bot doesn't appear in RLBot?

  • Double-check that the cpp_executable_path in CppPythonAgent.cfg is absolutely correct and points to the rlbot.exe file you compiled.

• Could not connect to server?

  • Ensure CppPythonAgent.cfg correctly points to your compiled rlbot.exe, and that the port numbers in rlbotmain.cpp match the one in GigaLearnCPP\rlbot\port.cfg

• Event Loop Stopped?

  • Make sure your port matches, if problem still occurs pip install websockets==12.0

Notes

• Ball prediction: Not supported. If your observation uses ball prediction, modify rlbotmain.cpp and pass a RocketSimArena to it. This is straightforward.

• Padded observations: Likely supported. To use, change:

auto obsBuilder = std::make_unique<AdvancedObs>();

to

auto obsBuilder = std::make_unique<YourObsNamePadded>(3);

Replace 3 with the value used during training.