Representation for Navigation

ABSTRACT: Matrix Research, Inc. proposes to prototype, test, analyze, and demonstrate novel feature representations in multi-agent simultaneous localization and mapping (MA-SLAM) systems for enhanced navigation accuracy/reliability. The objective of this effort is to develop a framework to integrate data from various air and ground sensor systems employed for exploitation and interdiction tasks, focused on compact, multi-modal feature representations tailored for navigation applications in which low-bandwidth, low-rate communications between agents constrains data throughput. A layered sensing processing framework is needed that combines information, gleans meaning from the collected data, and re-tasks sensor-seekers in a timely manner to garner actionable intelligence and execute a course of action. This is only possible if the platforms have sufficient context and are able to jointly localize and coordinate. Initially, this effort will focus on multi-view descriptors for wide-baseline correspondence. Then it will demonstrate these methods for joint localization and coordination to accomplish an objective. Our thesis is that tightly coupled"vision-based"navigation will provide the necessary foundation, context, and features for MA-SLAM. We have two transition paths lined up to commercialize our results, both of which contribute to the AFRL mission. These will include both indoor and outdoor navigation demonstrations. BENEFIT: The primary benefit of successful completion of this effort is a revolutionary new capability for extracting features for mapping systems. This capability has numerous commercial applications in various business sectors such as defense, search and rescue, mapping, mining, and robotics.