SBIR-STTR Award

To maximize situational awareness, while minimizing cognitive overload, automated abductive inference (best-explanation reasoning) will be used to create a changing, "best interpretation" representation of the situation from incoming data. Modeling and simulation will be used by the abductive inference software for automatic generation of predictions from hypotheses, enabling the continual generation of predictions to support: hypothesis evaluation, sensor tasking, planning, and detection of anomalies that may be valuable indications of deception, modeling errors, or sensor failure. Abductive inference will work tightly with predictive infererence to to provide a reliable, self-correcting representation of the situation, based on current evidence from sensor data, using domain knowledge encoded as causal-model fragments. Aetion proposes to extend its current technology base to create software for building and composing sensor-fusion applications that are modular and extensible as new types of sensors are integrated, and as new knowledge is available about object types, sensor characteristics, and causal processes that mediate the effects of objects on sensors. If this is feasible, the resulting software should be cost effective and highly valuable for multiple sensor fusion applications, resulting in systems able to squeeze more usable information from less data than systems not using causal relationships and domain models. Our automated inference technology is broadly applicable because it is based on a ubiquitous form of reasoning that is very human. So, development that benefits one particular application will tend to also expand our capabilities for others. As well as there being multiple customers with a need for advanced solutions in sensor fusion, our best-explanation approach shows exceptional promise for intelligence analysis, systems and situation monitoring, and diagnosis in engineering and medicine. Aetion's technology offers significant benefits for multiple aspects of the military's transformation over the coming decade, and we aim to realize that potential, by proving general and specific capabilities to the Department of Defense, then transitioning those products to address related commercial applications.

Keywords:
Sensor, Information, Inference, Reasoning, Data, Fusion, Explanation, Inference

To support pilot situation awareness for small high-speed submersibles in littoral environments, automatic information fusion is needed that is capable off deriving actionable information from sensor data that is degraded in the manner of sonar data in shallow water. A Phase I project showed the feasibility of using technology for automated abductive inference (best explanation reasoning), supported by modeling and simulation, to detect, and infer the locations of, sonically illuminated objects, using data from ambient sounds that have been degraded by multipathing. It appears that 3D scene reconstruction from passive (and active) sonar can be done using layered abductive inference, physics models, computational geometry, and miscellaneous domain knowledge. The objectives of the Phase II project are to develop a more advanced abduction engine and to demonstrate its effectiveness for 3D scene reconstruction from passive sonar. Model-based abduction has many potential applications to other problems of information fusion and interpretation for military and civilian applications.

Keywords:
SONAR, INFERENCE, SCENE RECONSTRUCTION, MULTIPLE HYPOTHESES, ABDUCTIVE INFERENCE, LITTORAL, NAVIGATION