| Summary: | Games are played for many reasons. It can be a platform for social interaction, a way
to challenge oneself, or just to escape reality [1]. Many games have a learning
curve and logical reasoning is usually required. Any opponents or enemies in a game
are usually hardcoded, thus they will be unable to match a human player since they
operating within limited and specific parameters.
With the introduction of newer deep reinforcement learning (RL) algorithms,
Artificial Intelligence (AI) will now do more than ever before, whether it was
playing a dynamic game like Dota or auto-generation of coherent images from partial
images, AI succeeded in all of these.
Without any experience in doing AI, it will be a great learning experience to be able
to train an AI for a strategic game without hardcoding it. Thus the objective of the
project is to build a imperfect information, multiuser online game integrated with AI.
Through this project, an AI was trained with a promising level of proficiency using
the Unity ML agent package, which simplifies the process of RL for a person who
does not have much knowledge about AI and its training, but yet able to use an
algorithm to train an RL AI for their purposes. If AI can be trained to human-like
proficiency with such a simplified interface, it will open up a whole avenue for the
future of games.
|
|---|