In recent decades, Unmanned Aerial Vehicles (UAVs), and quadrotor platforms in particular, have evolved from purely experimental research into commercial and industrial applications, offering rapid deployment, access to hazardous or remote locations, high-resolution imaging, and cost-effective operations. In this context, the present work addresses the need for high-performance platforms, reduced development time, and increased validation fidelity by proposing an integrated methodology that couples Software-In-the-Loop (SIL) and Hardware-In-the-Loop (HIL) testing strategies with Bayesian Optimization (BO), applied directly to the PX4 open-source autopilot. To this purpose, a high-fidelity, scalable model of quadrotors with varying mass and geometry is developed. The interfaces between the model and PX4 are implemented for both SIL and HIL configurations to enable high-fidelity closed-loop simulations. The autopilot tuning problem is solved via BO and executed in both SIL and HIL architectures. The resulting optimized controllers are compared with the nominal PX4 autopilot, and a Monte Carlo (MC) analysis is conducted to evaluate their robustness to model uncertainties and their performance relative to the nominal configuration. The results show an improvement in closed-loop performance compared with the nominal PX4 controller. Moreover, differences among the SIL and HIL simulations are highlighted, with the latter revealing system criticalities and leading to earlier actuator saturation. The principal contribution of this research is, to the author’s knowledge, the first application of BO for PX4 tuning within a HIL framework. This choice provides a higher fidelity level for the final controller by embedding real hardware limitations directly inside the optimization loop, and provides a validated testbed for future research on HIL-based data-driven controller optimization.
Negli ultimi decenni, i velivoli a pilotaggio remoto (UAV), e in particolare le piattaforme quadrirotore, sono passati dall’essere appannaggio di ricerche sperimentali ad applicazioni commerciali e industriali, offrendo minori tempi di impiego, accesso a zone pericolose o remote, acquisizione di immagini ad alta risoluzione e operazioni a costi contenuti. In tale contesto, il presente lavoro risponde alla necessità di sviluppare piattaforme ad elevate prestazioni, in tempi ridotti e con un livello di affidabilità sempre maggiore, unendo le strategie di test Software-In-the-Loop (SIL) e Hardware-In-the-Loop (HIL) all’ottimizzazione bayesiana (BO) applicata all’autopilota open-source PX4. A tale scopo, viene sviluppato un modello scalabile e ad alta fedeltà per quadrirotori con massa e geometria variabili. Vengono implementate le interfacce tra il modello e PX4, sia in configurazione SIL che HIL, per consentire simulazioni in anello chiuso. Il problema di tuning dell’autopilota viene risolto mediante BO e il loop di ottimizzazione viene eseguito in entrambe le architetture SIL e HIL. I controllori ottimizzati vengono poi confrontati con l’autopilota nominale di PX4, e viene condotta un’analisi Monte Carlo (MC) per valutarne la robustezza rispetto alle incertezze del modello e le prestazioni rispetto al controllore nominale. I risultati dimostrano miglioramenti nelle prestazioni in anello chiuso rispetto al controllore nominale PX4. Vengono inoltre evidenziate le differenze tra le simulazioni SIL e HIL, e queste ultime rivelano ulteriori criticità di sistema, portando a una saturazione prematura degli attuatori. Il contributo principale di questa ricerca è, a conoscenza dell’autore, la prima applicazione dell’algoritmo di BO per il tuning di PX4 utilizzando un’architettura HIL. Tale scelta garantisce un livello di affidabilità superiore per il controllore finale, integrando i limiti dell’hardware reale direttamente all’interno del ciclo di ottimizzazione, e compie il primo passo verso future ricerche sull’ottimizzazione data-driven di controllori basata su HIL.
Drone modeling and data-driven PX4 controller tuning via SIL/HIL
Cataffo, Francesco
2024/2025
Abstract
In recent decades, Unmanned Aerial Vehicles (UAVs), and quadrotor platforms in particular, have evolved from purely experimental research into commercial and industrial applications, offering rapid deployment, access to hazardous or remote locations, high-resolution imaging, and cost-effective operations. In this context, the present work addresses the need for high-performance platforms, reduced development time, and increased validation fidelity by proposing an integrated methodology that couples Software-In-the-Loop (SIL) and Hardware-In-the-Loop (HIL) testing strategies with Bayesian Optimization (BO), applied directly to the PX4 open-source autopilot. To this purpose, a high-fidelity, scalable model of quadrotors with varying mass and geometry is developed. The interfaces between the model and PX4 are implemented for both SIL and HIL configurations to enable high-fidelity closed-loop simulations. The autopilot tuning problem is solved via BO and executed in both SIL and HIL architectures. The resulting optimized controllers are compared with the nominal PX4 autopilot, and a Monte Carlo (MC) analysis is conducted to evaluate their robustness to model uncertainties and their performance relative to the nominal configuration. The results show an improvement in closed-loop performance compared with the nominal PX4 controller. Moreover, differences among the SIL and HIL simulations are highlighted, with the latter revealing system criticalities and leading to earlier actuator saturation. The principal contribution of this research is, to the author’s knowledge, the first application of BO for PX4 tuning within a HIL framework. This choice provides a higher fidelity level for the final controller by embedding real hardware limitations directly inside the optimization loop, and provides a validated testbed for future research on HIL-based data-driven controller optimization.| File | Dimensione | Formato | |
|---|---|---|---|
|
2026_03_Cataffo_Thesis.pdf
accessibile in internet per tutti
Descrizione: Thesis manuscript
Dimensione
18.12 MB
Formato
Adobe PDF
|
18.12 MB | Adobe PDF | Visualizza/Apri |
|
2026_03_Cataffo_Executive_Summary.pdf
accessibile in internet per tutti
Descrizione: Executive summary
Dimensione
2.69 MB
Formato
Adobe PDF
|
2.69 MB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/251989