Hybrid Machine Learning VDM for UAVs in GNSS-denied Environment

Abstract

This paper presents a novel approach to enhance unmanned aerial vehicle (UAV) autonomous navigation, without adding extra load to the vehicle. The proposed approach employs the UAV vehicle dynamic model to aid the navigation estimation in a Global Navigation Satellite Systems (GNSS)‐denied environment, without the need to model any part of the UAV, and avoids the requirement for special equipment during the modeling procedures typically required for vehicle dynamic model‐aided navigation. Taking advantage of the available information from previous flights during availability of GNSS, and with the aid of a hybrid machine learning approach, the proposed technique is able to enhance the navigation accuracy during GNSS outage, despite the massive drift occurring from utilizing a low‐cost inertial measurement unit (IMU) during the outage period. Different scenarios are investigated to prove the robustness of the proposed technique, and the results are compared to two types of IMUs with the aid of an inverse mechanization IMU simulator. © 2018 Institute of Navigation

Model-Based Autonomous Navigation with Moment of Inertia Estimation for Unmanned Aerial Vehicles

“…Zahran et al [32] investigates the use of hybrid machine learning to train a VDM and enhance inertial navigation accuracy during periods of GNSS outage using low-cost inertial sensors in a quadcopter. The machine learning module (regression and classification) acts as a substitute, providing a position and velocity solution during periods of GNSS outage.…”

“…The proposed architecture is preferred over the model-aided architectures [18,26,32] for a number of reasons. In low-cost applications, the quality of the inertial sensors used is relatively low and inherently affected by significant noise sources.…”

A Model-based Tightly Coupled Architecture for Low-Cost Unmanned Aerial Vehicles for Real-Time Applications

“…Also, they showed that the inclusion of the VDM parameters and wind in the state vector improved the stability of the filter and reduced error growth, especially in the presence of VDM parameter errors and wind. Zahran et al [13] derived a VDM from a hybrid machine learning scheme utilising a bagged regression and classification technique to aid an INS during a GNSS outage. The approach showed significant improvements in position estimation during an outage.…”

“…The vast majority of VDM integration schemes in the literature rely on using filtered GNSS measurements output by a GNSS receiver to provide a bounded navigation solution. Filtered measurements from a GNSS receiver are usually not available during a GNSS outage or when tracking less than four satellites, which can cause the navigation solution to drift even when using a VDM [9]- [11], [13], [16]. Further, model-aided INS schemes can easily be disabled in case of IMU failure and additionally, multi-process model schemes introduce duplicate states that increase computational cost [17].…”

A Drag Model-LIDAR-IMU Fault-Tolerance Fusion Method for Quadrotors

“…The state prediction of the dynamic model and INS are fused, and the procedure of the traditional Kalman filter is simplified. A dynamic model/INS/GPS fusion scheme was proposed by [23]. When GPS is available, the INS/GPS fusion results are used to identify the model parameters.…”