In the Machine Learning course we were tasked to develop an AI based on the Unity ML-Agents. To be able to finally beat my classmate at the famous game Blobby Volley, I decided to replicate the game in Unity and train an agent to play the game. This resulted in an agent I was unable to beat, as he just rarely let the ball drop on the ground.
The agent observes both players positions, his own velocity, the ball position and velocity and whether the ball is in the kick-off state. He is able to move left, right and jump. he gets rewarded whenever he hits the ball, plays it to the other side and when he scores. He gets punished when he jumps and plays the ball himself on the ground. He does not get punished when the other agent scores, as both are feeding the same brain.