Escape Protocol

Overview:

Escape Protocol is a reimagining of the game "Cops & Robbers," allowing the player to choose their desired role against AI opponents. Designed to be played in multiple rounds at a time, the player's goal is to simply complete their role's objective by outsmarting their enemies' constantly changing AI.

Gameplay Mechanics:

• Objectives - each role has a unique objective that they need to complete in order to win the round. The cops' objective is to tag all of the robbers before they can escape. The robber's objective is to collect all of the diamonds and return them to the purple deposit zone; robbers can only hold one diamond at a time.
• Abilities - each role has a special ability that they can activate to help them accomplish their objectives. Cops can temporarily double their movement speed. Robbers can temporarily hide themselves, preventing cops from tagging them. All abilities have a 10-second cooldown period.

Notable Features:

• Enemy AI Models - all enemies within the game are controlled through artificial intelligence models trained using the PPO (Proximal Policy Optimization) reinforcement learning algorithm. There are 20 models in total, 10 per role, and they were all trained in a variety of different environments to simulate various skill levels.
• Difficulty Evaluation Algorithm - at the end of each round, the game collects data on the player's performance, which is sent to an XGBoost algorithm to evaluate the current difficulty of the game; the result of this evaluation determines if the game's difficulty should be increased or decreased, subsequently changing the current AI models for the enemies. The data collected for this evaluation include the player’s role, success status, round duration, percentage of robbers tagged, percentage of diamonds deposited, and the amount of times either role activated their respective abilities.

Development Note:

This game was originally made for my college honors thesis, which examined the multiple appliciations of machine learning in game development. The inclusion of an algorithm to evaluate player performance was an attempt at creating dynamic difficulty settings as a method of removing difficulty as an accessibility barrier.

To handle the evaluation, the game uses a Python interpreter to load the algorithm model and return the prediction. Due to difficulties handling the interpreter with different builds, only a Windows build is currently available.