To Ask, Sense, or Share: Ad Hoc Information Gathering

Agents operating in complex (e.g., dynamic, uncertain, partially observable) environments must gather information from various sources to inform their incomplete knowledge. Two popular types of sources include: (1) directly sensing the environment using the agent's sensors, and (2) sharing information between networked agents occupying the same environment. In this paper, we address agent reasoning for appropriately selecting between such types of sources to update agent knowledge over time. In particular, we consider ad hoc environments where agents cannot collaborate in advance to predetermine joint solutions for when to share vs. when to sense. Instead, we propose a solution where agents individually learn the benefits of relying on each type of source to maximize knowledge improvement. We empirically evaluate our learning-based solution in different environment configurations to demonstrate its advantages over other strategies.