Particle filter based multi-sensor data fusion techniques for RPAS navigation and guidance

Abstract

This paper presents a Particle Filter (PF) based Multi-Sensor Data Fusion (MSDF) technique in an integrated Navigation and Guidance System (NGS) design based on low-cost avionics sensors. The performance of PF based MSDF method is compared with other previously implemented data fusion architectures for small-sized Remotely Piloted Aircraft Systems (RPAS). The sensor suite of the implemented NGS includes; Global Navigation Satellite System (GNSS) sensor, which is adopted as the primary means of navigation, Micro-ElectroMechanical System (MEMS) based Inertial Measuring Unit (IMU) and Vision-Based Navigation (VBN) sensor. Additionally, an Aircraft Dynamics Model (ADM) is used as a virtual sensor to compensate for the MEMS-IMU sensor shortcomings in high-dynamics attitude determination tasks. The PF is specifically implemented to increase the accuracy of navigation solution obtained from the inherently inaccurate, low-cost Commercial-Off-The-Shelf (COTS) sensors. Simulations are carried out on the AEROSONDE RPAS performing high-dynamics manoeuvres representative of the RPAS operational flight envelope. The Extended Kalman Filter (EKF) based VBN-IMU-GNSS-ADM (E-VIGA) system, Unscented Kalman Filter (UKF) based U-VIGA system and the PF based P-VIGA system performances are evaluated and compared. Additionally, an error covariance analysis is performed on the centralised filter using Monte Carlo simulation. Results indicate that the PF is computationally expensive as the number of particles is increased. Compared to E-VIGA and U-VIGA systems, P-VIGA system shows an improvement of accuracy in the position, velocity and attitude measurements.