SBIR-STTR Award

The proposed effort investigates machine learning and deep-learning artificial intelligence (AI) solutions to sensor degradation which is one of the more significant limitations that stymie reliable and dependable operations on the part of unmanned surface and undersea vehicles. Multiple external and internal causes related to adversarial action as well as system failure modes are investigated along with their effects on sensor system performance, and methods are developed to overcome these effects. A wide variety of navigation and tactical sensors and sensor functions are considered in exploiting overlaps in sensor coverage that may provide multiple perspectives to achieve sensor information redundancy by repurposing individual sensors in new roles for which they may not have been originally designed. Also considered is the introduction of new techniques that can compensate for degraded sensor performance through improved reasoning with higher levels of uncertainty present within sensor data streams. A third consideration is the exploration of the sensor signal characteristics of internal failures that distinguish themselves from external phenomena yet still manifest similar symptoms.

Benefit:
The proposed innovations apply directly to enhance U.S. Navy requirements to increase reliability, autonomy and expand USV/UUV mission capabilities for a wide variety of missions. These same capabilities may be applied to commercial markets related to Maritime Autonomous Surface Ships (MASS) that span offshore support vessels, hydrographic survey, fire fighting, law enforcement and short sea shipping.

Keywords:
mass, mass, autonomous vessels, Machine Learning, deep-learning, Sensors, Neural networks, Unmanned Ships, Artificial Intelligence

Sensor degradation over time is one of the more significant limitations that stymie reliable and dependable operations on the part of unmanned surface and undersea vehicles. The proposed Phase II technical effort provides a means to detect sensor degradation resulting from natural phenomena, adversarial action as well as internal vehicle failures. It also provides new ways to make adjustments needed to mitigate and compensate for reduced and degraded, corrupted and perplexing sensor inputs to accomplish mission objectives and facilitate mission success. Our approach involves sensor signal analytics through correlation in three different perspectives representing the distribution of signal characteristics across the frequency spectrum at any point in time, changes that take place in the signal over time and, where applicable, processed sensor signals as represented in imagery created from Frequency and Time domain signal data. Machine learning and deep-learning artificial intelligence are used to detect characteristic signatures and trends in sensor data indicative of rapid and gradual sensor degradation from various natural and manmade events. A Sensor Degradation Reasoning System examines a constant stream of external and internal vehicle sensor data, continually assess data characteristics for evidence of potential and actual degradation and, upon detection, recommend and implement alternative courses of action to maintain readiness and achieve mission objectives. This system is designed to be interoperable with onboard vehicle guidance, weapon and health monitoring systems to ensure the greatest chances of success in Navy end of Phase II testing and Phase III installation and deployment on Navy USV/UUV assets.