A Game-Contract Hierarchical Framework: Towards Autonomous UAV Swarm Collaboration in Uncertain Environments

UAV swarm collaboration based on dynamic hybrid architectures provides crucial technologies for intelligent networked systems. Existing studies struggle with local reactive strategies failing to balance coordination gains against communication costs, and the decoupling of high-level roles from low-level control hindering the translation of dynamic topologies into anti-oscillation physical actions. To this end, this paper proposes D-PACT, a game-contract hierarchical self-organizing coordination method that explicitly links topology partitioning, role assignment, and motion execution in communication-constrained environments. Macroscopically, a belief-based topology partitioning mechanism jointly considers future task gains, communication-maintenance costs, and energy expenditure to support stable decentralized topology adaptation. Microscopically, a coordination mechanism based on Graph Attention Networks and incentive-compatible contracts maps physical and communication capabilities into leader-member roles, which are further coupled with role-aware Model Predictive Control to translate organizational decisions into dynamically feasible physical motion. Experiments demonstrate that this method reduces mission completion time by 20.24% among completed runs and lowers the overall objective cost by 87.47% after excluding abnormal MILP objective outliers with JΣ>10.