Reinforcement learning to control a commander for capture the flag

Capture the flag (CTF) is a popular game mode for many blockbuster games. Agents in these games struggle against the players who learn to adapt to their strategies leading to the players dissatisfaction. We present our work on using Reinforcement Learning (RL) algorithms to learn a controller of a commander in the AI Sandbox platform, a flexible simulation environment which allows users across the world to participate in a variety of challenges and competitive games. As a result of building an RL controller for a commander we found that performance varies significantly across opponents, maps and team sizes, where the RL controller shows adequate performance in a subset of the games played and struggles in others.