Embodied Tactical Intelligence: Learning Sports Strategy Without Explicit Rules

Modern sports analytics systems rely heavily on hand-crafted rules, expert annotations, and predefined tactical concepts. While effective, such approaches fundamentally limit the emergence of generalizable intelligence. This paper proposes Embodied Tactical Intelligence (ETI)—a framework in which AI agents learn sports strategy directly from raw spatiotemporal observations and game outcomes, without access to explicit rules, labels, or domain knowledge. By framing sports as partially observable, physics-grounded, multi-agent environments, ETI enables the discovery of tactics as emergent phenomena rather than engineered constructs. We present a comprehensive learning architecture, propose rigorous evaluation metrics, and outline an experimental roadmap. Keywords: Embodied AI, Multi-Agent Reinforcement Learning, Sports Analytics, Emergent Behavior, Tactical Intelligence, AGI