Autonomous Decision Making via Hierarchical Brain Emulation -- 19-009

The objective of this project is to develop human intelligence-inspired algorithms that exploit multi-modal sources of low and high quality data to achieve a series of objectives such as detection, localization, tracking, and classification. A Bayesian model-based hierarchical adaptive decision making (HADM) algorithm will be developed which includes multiple levels of decision making organized in a hierarchical manner, a confidence factor associated with each decision, and a feedback mechanism used to trigger the need for higher quality data or to go back and correct erroneous intermediary decisions. A drawback of the Bayesian model-based approach is that the models required by the algorithm may not always be known or may be difficult to work with analytically. To overcome this limitation, an HADM algorithm that uses a model discovery-based approach to learn the required models from the data will also be developed using the exponentially embedded families approach to probability density function modeling and feature selection. Performance will be demonstrated on simulated radar and image data, as well as experimentally collected data from a laboratory testbed.