TOYON RESEARCH CORPORATION — Department of Defense SBIR Phase II: Toyon proposes to develop adversarial target prediction algorithms and a mission execution

TOYON RESEARCH CORPORATION — SBIR Phase II award from Department of Defense.

Amount
$730,000
Agency
Department of Defense · Army
Program / Phase
SBIR · Phase II
Solicitation
2010.2
NAICS
Place of performance
CA
Period
2011-12-22 → 2013-12-22

Description

Toyon proposes to develop adversarial target prediction algorithms and a mission execution framework that supports collaborative teaming of airborne and ground-based Unmanned Systems (UMS) to enable tracking and surveillance of evasive targets. The proposed AVERT system will include automatic video processing, global fusion and tracking, adversarial reasoning and prediction logic, automatic UMS feedback control, and an Android-based application for a handheld device. Using Toyon's custom GeoTrack embedded hardware, the video processing and global fusion and tracking capabilities will exist in parallel on-board the UMS and on the ground station, adding flexibility to the system architecture. The on-board processing enables AVERT to operate with a decentralized architecture; each UMS can operate autonomously, but a wireless network can support data sharing for collaborative actions between several UMS. The on-board functionality will also enable the Android app and device to receive video, telemetry, and track data directly from all UMS within radio range, without hopping through a ground station. The handheld operator can also input ISR requests to the handheld that will be processed by the UMS. Toyon's field-proven video processing algorithms will automatically detect moving targets in video. Target detections will be automatically geo-registered using sensor state data from the platform GPS and IMU, then fused into target tracks in the ground plane, enabling autonomous tracking even in urban and complex environments. Ground plane tracks will be sent to the UMS operator for viewing, and passed on to UMS control algorithms. Our control algorithms will use adversarial modeling to anticipate potentially evasive actions of the target, then optimize the platform trajectory and sensor orientation to obtain clear-line-of-sight to the target and maintain persistent surveillance.