Marcel Založnik and Kristjan Šoln

Abstract
This paper presents a multi-agent system (MAS) for autonomous table football, developed for the FuzbAI competition at the University of Ljubljana. Our system consists of four independent agents, each dynamically assigned specific roles—Goalkeeper, Defender, Midfielder, and Attacker—based on real-time game analysis. This role-based architecture enabled seamless coordination between offensive and defensive strategies, allowing our team to secure first place. We describe the simulation framework used, the processing of sensor data, and the control strategies that allowed the agents to execute precise actions in a dynamic environment. The results highlight the effectiveness of adaptive, role-based decision-making, demonstrating the potential of MAS in real-time, competitive settings.