Decision Making in Monopoly Using a Hybrid Deep Reinforcement Learning Approach

Learning to adapt and make real-time informed decisions in a dynamic complex environment is challenging problem. Monopoly popular strategic board game that requires players multiple during the game. Decision-making involves many real-world elements such as strategizing, luck, modeling of opponent’s policies. In this paper, we present novel representations for state action space full version define an improved reward function. Using these, show our deep reinforcement learning agent can learn winning strategies against different fixed-policy agents. Monopoly, take actions even if it not their. turn roll dice. Some these occur more frequently than others, resulting skewed distribution adversely affects performance agent. To tackle non-uniform actions, propose hybrid approach combines (for frequent but decisions) with infrequent straightforward decisions). We develop agents using proximal policy optimization (PPO) double Q-learning (DDQN) algorithms compare standard proposed approach. Experimental results outperform by 20% number games won The PPO performs best win rate 91%