Foosball is a fast-paced, strategy-driven table game that requires sub-second decision-making, fine motor control, and dynamic tactics. Reinforcement Learning (RL) algorithms are widely used for AI agents to implicitly learn the physical world. In this work, we present a new open-source framework designed for evaluating end-to-end deep RL algorithms in a simulated foosball environment. Our platform includes a high-fidelity physics simulation environment built in MuJoCo, enabling analysis of performance transfer from virtual to real-world conditions. This platform is designed to advance the development of competitive agents capable of robust, adaptive behavior and effective sim-to-real generalization, serving as a standardized resource for the broader RL community.