Multi-Agent Reinforcement Learning-Implementation of Hide and Seek

Multi-agent reinforcement learning (MRL) is the study of several agents (learning and acting part of the problem) living together in an environment to achieve the end goal. Using a reinforcement algorithm, the aim is to optimize the future rewards by predicting the appropriate action of the agent for reaching goal state. We have applied this theory to implement a simple game of hide and seek, where a team of seekers (multiple agents) would train themselves to seek a team of hiders and vice versa. The objective of the research paper is to study various reinforcement theories and algorithms like Temporal Difference Learning, Q-Learning, SARSA and select/choose an appropriate algorithm to implement a game of hide and seek. This paper also explains how we implemented the hide and seek game, with multiple agents and the insights that we came up with after observing the various episodes of the game in the environment.