Building your kingdom Imitation Learning for a Custom Gameplay Using Unity ML-agents

Imitating human-like behavior in action games is a challenging task in machine learning research, with several methods being employed to solve that problem. As individual players often have characteristic styles when playing the game, this method attempts to find these behaviors which makes it unique. In this paper, we consider learning human-like behavior via Unity ML-Agents Toolkit without being explicitly given a reward function, and learning to perform the task by observing expert's demonstration. The use of imitation learning makes every player imprints his behaviour on the clone(Agent).Therefore, each agent will play differently because they are following the main player's(Teacher) patterns. Our work shows there is significant reduction in computational time and the amount of training data needed by using imitation learning on top reinforcement learning.