Joint optimization of detection, task allocation, and path planning for multi-agent cooperative interception under partial observability

Safety-critical autonomous systems often require multiple agents to detect, assign, and intercept non-cooperative mobile targets using only partial observations. These subtasks are tightly coupled because the same agents must balance spatial exploration for detection with goal-directed motion for interception, while target maneuvers remain uncertain. This study develops and validates an integrated multi-agent coordination framework for jointly optimizing distributed detection, dynamic task allocation, and path planning under partial observability. The problem is formulated as a coupled Dec-POMDP made computationally tractable through factored decentralized policies, parameter sharing, and centralized training with decentralized execution, and a Weighted Coordinated Multi-Agent Proximal Policy Optimization (W-MAPPO) framework is introduced to adaptively coordinate competing subtasks. W-MAPPO combines an observation-conditioned Policy Ratio Predictor, an LSTM-based history-aware processor for temporal modeling of target maneuvers, and a Voronoi-based collaborative detection module for spatial coverage coordination within a centralized-training decentralized-execution paradigm. Comprehensive simulation experiments across increasing target densities compare W-MAPPO with seven representative MARL baselines and two classical methods, supported by statistical testing, ablation studies, sensitivity analysis, and representative case studies. W-MAPPO achieves a 94% success rate in the 5 vs 5 scenario and maintains stable interception performance as the number of targets increases to 10 and 15. These results demonstrate the technical reliability and practical potential of integrated subtask coordination for partially observable multi-agent systems.