VoI-Driven Multi-UAV Collaboration for Data Collection With Partially Outdated Observation

This letter proposes a Value of Information (VoI)-driven multi-Unmanned Aerial Vehicle (UAV) collaboration framework for Internet of Things device (IoTD) data collection in unknown environments. Each UAV plans its path using partially outdated local observations shared among neighbors via the proposed knowledge-sharing mechanism under limited communication. To address partial observability and inter-agent dependency, we propose a Multi-Agent Transformer-based Proximal Policy Optimization (MATPPO) algorithm, in which the critic network is replaced with a Transformer encoder to capture the interactions between UAVs and IoTDs. Experimental results indicate that MATPPO achieves superior performance in terms of the collected VoI compared to traditional baselines.