GPS DENIED NAVIGATION USING LOW-COST INERTIAL SENSORS AND RECURRENT NEURAL NETWORKS

Abstract

Autonomous missions of drones require continuous and reliable estimates for their velocity and position. Traditionally, Extended Kalman Filtering (EKF) is applied to measurements from Gyroscope, Accelerometer, Magnetometer, Barometer and GPS to produce these estimates. When the GPS signal is lost, estimates deteriorate and become unusable in a few seconds, especially when using low-cost inertial sensors. This thesis proposes an estimation method that uses a Recurrent Neural Network (RNN) to allow reliable state estimates in the absence of GPS signal. On average, EKF positioning error grows to around 40 kilometers in five minutes of GPS-less typical drone flight. The proposed method reduces that error by 98% in the same GPS