You are here

Decision Making under Uncertainty


OBJECTIVE: Analyze the impact of sensor measurement uncertainties on centralized data fusion and design optimal strategies to mitigate the associated target classification. DESCRIPTION: This topic solicits innovative approaches to characterize target sensor measurement uncertainties and to design effective sensor architectures to aid uncertainty mitigation (e.g. whether sending measurements or tracks between platforms enables uncertainty mitigation more effectively). Proposals are sought to research origins of uncertainty from sensor measurement to processing at a centralized command and control, C2, node for target classification. Novel techniques should be developed to effectively quantify and manage that uncertainty with respect to classification. Investigation should be conducted into measurement uncertainty associated with decision level fusion as compared with measurement level fusion at C2 for two reporting sensors. Currently the industry state of the art is to convolve the uncertainty distribution with the probability distribution or to express the probability distribution conditioned on the measurement uncertainty. One approach has been to propagate the uncertainty through the fusion chain. However, these types of analyses have not been adequately considered as a link in the data fusion chain and how the incorporation, and where in the processing chain, the uncertainty is incorporated effects the classification decision, and the cumulative uncertainty with respect to the final decision. Each individual sensor that detects, tracks, and takes measurements on the target could pass measurements, features or classification decisions to C2BMC. This functional architecture affects the quality and uncertainty of the fused classification result. The initial effort should focus on two sensors, one infrared (IR) and one radar (RF), observing a target and the classification function and its uncertainties. The approach should investigate measurement or feature level fusion (measurements or features received from the various sensors), and compare that with decision level fusion (decisions sent to C2 by the sensors). Follow-on efforts can incorporate the uncertainty associated with tracking errors and track correlation uncertainties. The goal of this effort is to research how target uncertainties propagate through various fusion methodologies. PHASE I: Develop and demonstrate a method to capture target characterization uncertainties from data received from two sensors observing a target simultaneously and fusing the information to make a classification decision. Compare the uncertainties with decision level fusion. Evaluate the classification output from the C2 node. PHASE II: Refine and update concept(s) based on Phase I results, and incorporate the added uncertainty from tracking and track correlation in the presence of multiple targets and non-simultaneous observation. Demonstrate how the target classification decision by C2BMC can be characterized with respect to the accumulating uncertainties in the system and demonstrate methods to reduce that uncertainty. A government testbed will be made available at no cost to the proposing firm to coordinate high fidelity testing. PHASE III: Demonstrate the new technologies via operation as part of a complete system or operation in a system-level test bed to allow for testing and evaluation in realistic scenarios. DUAL USE/COMMERCIALIZATION POTENTIAL: The contractor will pursue commercialization of the various technologies, analyses and design components developed in Phase II for potential commercial and military uses in many areas such as automated processing, robotics, medical industry, and in manufacturing processes. This could be valuable in mission critical decision making systems like automated diagnostic systems, or alarm systems where false alarms can be costly. In automated processing and manufacturing, it could apply to quality control.
US Flag An Official Website of the United States Government