The research objective of this award is to derive, develop, implement and test a theoretical framework for control and estimation methods for sensor networks that enables the derivation of near optimal rates of approximation of the vehicles' environment. The theoretical framework will serve as the foundation for deriving encoding and decoding schemes that are essential to high-bandwidth mobile sensor networks. The approach relies on the synthesis of techniques from several disparate technical fields including control theory, estimation theory, approximation theory and statistical learning theory. The work extends recent efforts that study the fidelity of approximation in distribution free learning theory, a sub-problem of statistical learning theory, to include classes of decentralized and dependent measurement processes. The algorithms developed under this project will be validated using multi-vehicle network that host three dimensional laser-ranging sensors to map unknown environments.
The successful completion of the research will enable novel and efficient algorithms for the mapping of unknown fields over complex environments by decentralized, autonomous robotic vehicle teams. The successful completion of the research will enable large-scale, multivehicle sensor networks to be employed in a host of environmental monitoring and mapping applications including contaminant dispersal in a marine environment, chemical plume dispersal in urban environments, wildfire evolution in forests, and crop density in automated agribusiness. This research project will build human and institutional infrastructure for science and technology via the creation of new graduate courses that treat the synthesis of estimation, control and approximation theory. The research program will create an outreach program that engages K-12 students in robotics and environmental mapping. The results of this research will be disseminated through presentation and publication at national and international conferences, as well as through top-notch peer-reviewed journals. The dissemination of the research will enable industry to implement and field scalable and general algorithms for decentralized mapping using robotic vehicle sensor networks.
