Published Paper Details:

This study presents an interactive, real-time simulation combining deep reinforcement learning and adversarial dynamics in a multi agent treasure maze setting. Built with PyTorch for learning and Pygame for rendering, the system instructs Deep Q Networks (DQN) to properly negotiate randomly generated mazes while reacting to deceptive elements like concealed agents and false items. Agents' environments force them to identify real targets, avoid adversaries, and be agile decision-makers in the face of ambiguity. Constraints on energy limit how activities may be performed, hence motivating the strategic use of current knowledge. Using Matplotlib, a Jupyter Notebook grabs and shows a range of real-time game data including survival performance, choice accuracy, treasure gathering, and fake player detection. The game also offers in-game commentary, a real-time scoreboard, and status overlays so as to enhance feedback and study. Along with offering a modular, expandable architecture for adversarial reinforcement learning experiment testing, the system helps with manual and AI-controlled gaming