Learned Memory Attenuation in Sage-Husa Kalman Filters for Robust UAV State Estimation
Teaching drones to track themselves better when sensors fail
Drones lose track of their position when sensors cut out or vibrate unpredictably—problems that stumped earlier tracking systems. Researchers built a learning-based filter that adapts to these disruptions in real time, using a neural network to adjust how much it trusts past measurements versus new sensor data. On real drone flights, it stayed accurate longer than standard methods when sensors went dark.
Drones operating in cluttered or noisy environments—industrial inspection, search and rescue in cities, GPS-denied zones—depend on reliable position estimates to avoid crashing. This filter extends how long a drone can navigate safely without external signals, and keeps it oriented during the messy transition when it must switch from sensor data to pure dead reckoning. That directly improves safety and mission success in real-world conditions where classical filters fail.