KATANA: A Fast, Low-Power Mapping of Kalman Filters onto Edge NPUs for Real-Time Tracking
Researchers have developed KATANA, a novel framework that enables the efficient execution of Kalman Filters on Neural Processing Units (NPUs) found in modern AI-PC SoCs. This approach aims to overcome the limitations of using traditional CPUs or custom hardware for real-time tracking systems, which are critical in applications like autonomous driving and defense. By optimizing the Kalman filter algorithms for NPUs, KATANA achieves significant improvements in speed and power efficiency, freeing up the CPU and GPU for other tasks. AI