Enhanced Scene Reasoner

To address the Army’s need for an automated semantic reasoner capable of using logical rules to deduce a local world model from a changing knowledge base consisting of multi-modal data that also utilize online learning techniques to adapt to changing environments, Physical Optics Corporation (POC) proposes to develop a new Enhanced Scene Reasoner (ESR) algorithm suite. It is based on a new system design that incorporates relation discovery, relation reasoning, and novel use of an architecture to learn from one unknown class only. This enables ESR to locate and identify entities of a scene, determine and reason relations between any two entities, and evaluate and display correctness of classification based on its reasoning. ESR will report results to a user who provides man-in-the-loop feedback to improve the reasoner’s and classifier’s performance. The ESR suite of algorithms offers a 100x reduction in rate of incorrect classification of prominent objects and regions compared to a baseline, which directly address the Army’s Phase III Next Generation Combat Vehicle and Soldier Lethality modernization requirements. In Phase I, POC will demonstrate the feasibility of ESR by reaching technology readiness level (TRL)-3. In Phase II, POC plans to develop an ESR prototype algorithm suite reaching TRL-6.