Physical inactivity and the lack of meaningful social interaction remain pressing challenges in the increasingly digital environments of today. While exergames offer a promising avenue to encourage movement, they often lack real-time adaptability and social engagement mechanisms. However, few existing systems integrate real-time Machine Learning (ML)-based activity recognition with peer-assisted interaction to enhance both physical and social dimensions of gameplay. We present SpaceStriker: Exercise Odyssey, a real-time sensor-based Exergame that combines ML-driven feedback with structured peer interaction modes to promote collaborative and competitive play. The system uses real-time motion recognition to provide immediate feedback, supporting fairness, engagement, and co-regulation between players. Evaluation of user feedback, behavioral observation, and gameplay data across multiple roles reveal that participants found the system usable, motivating, and socially engaging, with peer assistance enhancing awareness and accountability. Our findings demonstrate the feasibility and potential of integrating peer collaboration and ML-based sensor feedback in exergames, paving the way for scalable and, adaptive applications in educational, therapeutic, and social settings.