Learning Pose Estimation for UAV Autonomous Navigation andLanding Using Visual-Inertial Sensor Data

TL;DR

This paper introduces a deep learning-based multimodal approach for UAV pose estimation using visual-inertial data, demonstrating improved accuracy and successful autonomous navigation and landing in simulation.

Contribution

A novel end-to-end neural network architecture that fuses visual and inertial data for UAV pose estimation, outperforming traditional methods.

Findings

25% improvement in pose estimation accuracy over baseline methods

Successful integration into a flight control system for autonomous navigation

Effective simulation results for UAV landing and navigation

Abstract

In this work, we propose a new learning approach for autonomous navigation and landing of an Unmanned-Aerial-Vehicle (UAV). We develop a multimodal fusion of deep neural architectures for visual-inertial odometry. We train the model in an end-to-end fashion to estimate the current vehicle pose from streams of visual and inertial measurements. We first evaluate the accuracy of our estimation by comparing the prediction of the model to traditional algorithms on the publicly available EuRoC MAV dataset. The results illustrate a $25 \%$ improvement in estimation accuracy over the baseline. Finally, we integrate the architecture in the closed-loop flight control system of Airsim - a plugin simulator for Unreal Engine - and we provide simulation results for autonomous navigation and landing.