Implementasi Game Tag dengan Machine Learning Menggunakan Unity

Abstract

The implementation of games using machine learning is very rarely discussed and discovered, but its development is very rapid because it is digital-based, therefore research is needed that discusses the combination of machine learning and games. This game will be created using Unity which is already a program with the Unity machine learning agent. In this research, a tag game was built using a highly extensible Unity machine learning agent. Unity machine learning agent uses the proximal policy optimization (PPO) algorithm which is classified as a policy gradient method to train the agent policy network. In the implementation of this tag game, it is a game that has a minimum of 2 or more players, where 1 will be the catcher and the other player will be the runner. In this game, we see how efficiently the enemy agent can be trained with the variable average remaining time so that it can create a strategy to achieve the specified goals as quickly as possible and get the best generation. In this game, testing was carried out to see the reliability of the Unity Machine Learning Agent in capturing players from the 1st generation to the 5th generation. The results obtained until testing the 5th generation are that the 3rd generation has the worst time, namely 70 seconds with an average remaining time and the best is the 5th generation with an average remaining enemy agent time of around 252 seconds from a game time of 300 seconds. Then the 5th generation had a value of 0 failed arrests for users, so research was carried out until the 5th generation because it was considered reliable in playing with players.