Research objectives and approaches. The objective of this research is to develop probabilistic methods for active sensing with tunable chemosensors. The approach is formulated as a partially observable Markov decision process, whereby the system takes sequential sensing actions to identify a chemical target. The investigators will develop algorithms to adapt sensor tunings on-the-fly for two types of scenario: target discrimination and background rejection. These algorithms will be validated on two tunable sensor technologies (metal-oxide chemoresistors and infrared interferometers) combined with tunable preconcentrators. Intellectual merit. Active sensing has received minimal attention in chemosensors, where most work has focused on off-line optimization. Likewise, chemometrics techniques assume that a full spectrum of the chemical is available. In contrast, the proposed work argues that only the most informative wavelengths/features are needed, and that these should be selected dynamically. This research will lead to agile systems that can adapt sensing parameters on-the-fly in response to their environments. The work will also provide a better understanding of active sensing across a range of scenarios (e.g. static vs. dynamic stimuli) and sensor characteristics (e.g. specific vs. cross-selective). Broader impacts. For a number of applications, from quality control to environmental monitoring, there is a need to rapidly detect, identify and quantify volatile compounds. The proposed methodology generalizes to a large class of chemical detectors, and may enable the development of power-aware networks of chemical sensors. The project will provide research opportunities for undergraduate students; since 2001, the PI has supervised 45 senior-design projects involving over 140 students.
