A Research Framework for Asynchronous Adversarial Multi-Player Games with Human Player GUI and AI Gym
Type of DegreeMaster's Thesis
Computer Science and Software Engineering
MetadataShow full item record
The primary contribution of this thesis is to describe the design and development of a research framework to support complex, real-time asynchronous multi-agent simulations with both homogeneous as well as heterogeneous human and artificial intelligence (AI) agents. This framework provides researchers with a platform to model adversarial games and benchmark AI algorithms and policies. It provides a flexible, reusable client/server architecture that supports a wide variety of games for use as (learning) environments. The secondary contribution of this thesis is to describe the design and development of an example use-case of the research framework for satellite constellations named Satellite Tycoon (Sat-Tycoon). The provided React client provides researchers, educators, and aerospace enthusiasts with a way to play the provided Sat-Tycoon game from their internet browser without the need for downloading or installing. The Sat-Tycoon game itself is a satellite constellation simulation rich with complexity, ideal for use as a reinforcement learning environment. This thesis will cover the design decisions made for these contributions, and the trade-offs involved in those decisions. Such decisions include using multiple networked gyms instead of a single multi-agent gym, using the OpenAI gym standard instead of PettingZoo, building a server authoritative architecture over a peer-to-peer architecture, and using a web framework instead of a game engine.