Unmanned Aerial Vehicles (UAS)
For military drones to work their best, artificial intelligence and machine learning solutions have to run in extreme conditions—on the edge, in the sky, and on low-power hardware. Understand how we right-size our network architectures and deploy our solutions to low-size, low-weight, and low-power systems.
- Artificial Intelligence, Machine Learning, and Knowledge Exploration
- U.S. Army and NVESD
Big Changes in Small Packages
Tiny drones have modernized how we gather intelligence and fight enemies. With our partner U.S. Army Night Vision and Electronic Sensors Directorate (NVESD), we have placed the power of artificial intelligence and machine learning target acquisition onto these small marvels of technology. Now our warfighters can know what enemies are around the bend—before those enemies know we are there.
How Big Data Works on Small Unmanned Aerial Vehicles (UAS)
Our algorithms for drone-based target recognition maintain a perfect balance of cutting-edge artificial intelligence and machine learning and low-size, low-weight, low-power software deployments. Our engineering solutions ensure that networks are right-sized, are compiled to use less power, and run at the required frame rates.
In an ongoing SBIR, we matured our DOCTRINAIRE software to run in real-time, on low-power hardware (e.g., NVIDIA JETSON). It performed robust aided target recognition (AiTR) at a fraction of the computational complexity typically needed for deep-learning computer vision systems.
Our world-space target localization uses the optimal combination of GPS, Inertial Measurement Units (IMU), and image data to provide stabilized target-declarations that are compatible with TAK and other down-stream targeting systems.