I'm training an AI bot to play Rocket League with GigaLearnCPP, a C++ reinforcement learning framework built on RLGym-PPO. Instead of hardcoding if-this-then-that rules, the bot plays itself thousands of times and learns from a reward signal what actually wins games.

How it works

• Reinforcement learning (PPO) — the bot is rewarded for things like scoring, ball touches and boost control, and adjusts its policy over many games
• GigaLearnCPP runs the training loop in C++, which is far faster than the Python-based RLGym-PPO it's built on
• The trained model gets loaded into RLBot so it can actually play matches against other bots or players

Pulled straight from the training run logged in Weights & Biases.

As of Aug 11, 2026.

Status

Still training and tuning — this page will get a highlight clip once the bot is consistently beating scripted opponents.