Miccio, Enrico (2025) ADVANCING AUTONOMY IN AVIATION: OPTICAL AND RADAR SENSING PIPELINES FOR AI-POWERED APPROACH AND LANDING AND ENHANCED GROUND SURVEILLANCE. [Tesi di dottorato]

Documento PDF miccio_enrico_38.pdf Visibile a [TBR] Amministratori dell'archivio Download (8MB) | Richiedi una copia

Abstract

The aviation industry has experienced major changes over the last decades. These transformations are driven not only by increasing air traffic and the consequent need for higher automation, but also by the technological progress and the emergence of new markets requiring autonomous platforms. Such advances are enabling new concepts such as air operations in urban environments – commonly referred to as Urban Air Mobility (UAM). The push toward more automated and fully autonomous flying platforms demands huge effort to overcome numerous new challenges. To face some of the challenges, this dissertation proposes different solutions employing perception systems integrated into complex processing pipelines to support both manned and unmanned operations. Specifically, a complete radar-vision-aided autonomous navigation pipeline is presented to support autonomous approach and landing operations within UAM scenarios. While a standalone radar-based solution is proposed to enable independent airport surfaces surveillance to mitigate incidents such as ground collisions or runway incursions, which stand as two of the most critical issues. The autonomous navigation pipeline for UAM scenarios is necessary since conventional navigation systems available at airports such as typical GNSS-based systems may be inapplicable or unreliable in complex urban environments. The proposed solution integrates two visual cameras, a Frequency Modulated Continuous Wave (FMCW) radar, an Inertial Measurement Unit (IMU) and a GNSS receiver. And it is tailored to work in accordance with preliminary regulations and designs guidelines established by the European Union Aviation Safety Agency (EASA), even if it is meant to be adapted to different cases. By providing autonomous capabilities to the vehicle itself, the system reduces the necessity of having complex ground infrastructures and emphasizes the autonomy of the aerial platform. The vision-aided part also comprehends an obstacle detection module capable of detecting incursions at landing and obstacles on the landing area. This vision-aided module relies on Artificial Intelligence (AI)-based algorithms, which offer generalization and provide higher robustness and reliability if compared to traditional computer vision techniques. The methodology requires the detection of the landing area, which is performed by a dedicated Convolutional Neural Network (CNN) trained for the specific operational domain. Then, custom motion-aware image processing methodologies are used to detect, match, and track visual landmarks which are then used to solve a perspective-n-point problem through an iterative non-linear least squares strategy. The radar-aided module exploits ad-hoc algorithms that leverage standard techniques. It is designed to detect high-reflectivity targets placed on the landing area and uses a matching algorithm to associate these detections with their known positions, enabling reliable corrections to the aircraft navigation state. The integration of all the sensors is made by means of an Extended Kalman Filter (EKF). The visual measurements are loosely integrated into the filter, while radar data is tightly integrated into the EKF so that the filter can exploit even a single radar target association to correct the state. Analysis of residuals, as well as covariance-based checks on the difference between predictions and estimates, are used before feeding correction solutions into the filter, thus providing protection against outliers. Performance assessment is first carried out by means of numerical simulations within high-fidelity virtual environments able to realistically reproduce aircraft dynamics, sensors’ operation, illumination and weather conditions. Such analysis is then complemented by experimental tests conducted with a small unmanned aerial vehicle. The airport surface surveillance application exploits, on the other hand, a FMCW radar operating in K-Band but in this case the system is, at the current stage, analysed as a standalone solution independent from external sources of information. The methodology adopted starts from the raw sensor data up to the generation of confirmed tracks, which are then georeferenced in order to compare them to benchmark data provided by the ADS-B system. The pipeline leverages well-established algorithms, tailored to the specific characteristics of the scenario under investigation. The sensor together with the proposed methodology has been validated through flight-testing data collected at a US-based airport across various operational scenarios.

Downloads

Actions (login required)

Modifica documento