basketball-multiagent-hips

# Performance Optimization of Multi-Agent Cooperative Algorithms in Basketball Tactical Simulation ## OverviewThis repository implements the research described in the paper **"Performance Optimization of Multi-Agent Cooperative Algorithms in Basketball Offensive and Defensive Tactics Simulation"**. The core focus is the development of **Hierarchical Policy Synchronization (HiPS)** within a **Cooperative Decision-Making Network (CDMN)** framework to optimize multi-agent cooperation in basketball tactical simulations. ## Key Features- **Hierarchical Policy Synchronization (HiPS)** for scalable multi-agent coordination. - **Cooperative Decision-Making Network (CDMN)** with: - Structured Global-Local Policy Learning - Robust Communication and Coordination - Adaptation to Dynamic Agent Populations - Improved convergence speed, scalability, and cooperation efficiency. - Benchmarked against state-of-the-art methods (SlowFast, TimeSformer, VideoMAE, etc.) with superior performance results (e.g., **91.76% accuracy** on NSVA dataset). - Includes **Resilient Learning & Exploration module** for fault-tolerant learning under partial synchronization. ## Results SummaryAccording to the paper: - Achieved **23% improvement in coordinated success rate** and **17% reduction in reaction time** compared to baselines. - Outperformed SOTA methods across multiple datasets (NSVA, Video-MME, Aesthetic Visual Analysis, OpenSpiel). - Ablation studies confirm the importance of each component (Structured Policy Learning, Communication, Hierarchical Group Coordination). ## Repository Structure