Assignment-Consistent Dynamic Multi-UAV Task Allocation: Communication-Efficient Dynamic Multi-Unmanned Aerial Vehicle Task Allocation Under Stale and Asymmetric Information

Dynamic task allocation for unmanned aerial vehicles (UAVs) is often constrained by stale and asymmetric information. Variants of the Consensus-Based Bundle Algorithm (CBBA) can waste bandwidth by transmitting every state or bid update without distinguishing assignment-altering information from irrelevant perturbations. This paper proposes Assignment-Consistent Dynamic Multi-UAV Task Allocation (AC-DMTA), a decision-centric framework in which agents communicate only when local uncertainty may invalidate the current assignment. AC-DMTA represents stale records with uncertainty sets and converts point bids into interval bids, enabling task-winner and bundle-stability certificates to suppress assignment-irrelevant messages. These certificates provide local sufficient conditions for certifying the stability of task-winner and bundle decisions within the modeled uncertainty envelopes and candidate sets, allowing assignment-irrelevant communication to be skipped. Timestamp and recency records further resolve asynchronous conflicts without broad resynchronization. Across four benchmark environments and 20 paired seeds, AC-DMTA sends fewer bytes and directed messages than periodic CBBA, Event-Driven Consensus-Based Bundle Algorithm (ED-CBBA), Clustered CBBA, and Two-Level Clustered Consensus-Based Bundle Algorithm (TLC-CBBA), achieving byte reductions of 10.4–91.1% across baseline–scenario pairs while maintaining competitive reward and task-completion rates. Under the tested delay and packet-loss stress conditions, AC-DMTA also achieves the lowest number of bytes per completed task among the compared methods.