Coordinating Learning Agents for Active Information Collection
Kagan Tumer · 2011
Abstract : The ability to autonomously coordinate a team of agents to actively collect information is critical to a wide array of Air Force missions. With computing power becoming both cheaper and more powerful, there is a trend to push critical decision making capabilities downstream, towards the data collection nodes rather than wait for data to arrive to a massive centralized location before a decision is made. This new computing paradigm relies on networked agents to actively collect, process and query data and promises to significantly improve both the quality/relevance of the collected data and the associating decision making. This project provides a comprehensive solution to the problem of intelligent data gathering and decision making by ensuring that the information collected by an agent has the most added value to the full network. The key contribution of this project is to shift the focus from how to optimize to what to optimize in difficult coordination problems. The impact of this work extends to a large class of problems relevant to the Air Force inlcding satellite communication systems, reconfigurable flight control systems, sensor networks, and intelligence gathering in hybrid networks.