GTC Silicon Valley-2019 ID:S9216:Using GPUs in the Development of Airborne Collision Avoidance
Sheila Jaszlics(Pathfinder Systems, Inc.)
Learn how we used GPU technology to integrate Unmanned Air Systems (UAS) into the national airspace. By using a suite of miniature, high-performance video cameras and deep neural nets, the technology implements a design for a 360-degree detect-and-avoid capability. The passive nature of the solution greatly reduces system power and weight, making the technology applicable to a wide array of UAS and piloted aircraft at a low cost. We'll describe how the technology was initially developed for remotely piloted aircraft but is equally applicable to small commercial or military systems, and will soon become affordable even for the below 55-pound class. Our talk will also cover critical challenges we faced, including determining the optimal use of CNN in collision avoidance, sensing and modeling the host aircraft within an air mass, real-time dynamic flexing of the host aircraft, and the creation of intruder aircraft models based on CNN recognition.