Autonomous racing represents the ideal scenario to boost autonomous driving development and testing, as it pushes the limits of technology through competition, while remaining confined to a controlled and regulated environment. In recent years, self-driving competitions such as Indy Autonomous Challenge, Roborace and Formula Driverless are becoming popular. In this context, several challenges arise, requiring more advanced and performant control systems for sports and racecars and resulting in novel solutions for vehicle dynamics control at the limits of handling. In this thesis, we focus on the development and experimental validation of feedback control systems for high-performance vehicles in sport driving conditions, including strongly model-based control techniques such as Model Predictive Control. The research is based on the traditional architecture with steering and throttle/brake control inputs, with an extended high-speed experimental validation on a full-scale racecar. Besides that, we explore the potentialities of alternative architectures that are common in the world of high-performance vehicles. We focus on torque vectoring (for electric and hybrid vehicles with multiple electric motors) and four-wheel-steering. Despite these latter applications are limited to the simulation environment, attention is placed on the limitation of computational complexity for potential real-time implementation on low-power hardware. In the development of the control strategies, it is fundamental to deal with several side aspects of vehicle dynamics control. In particular, we cover the design of hardware-in-the-loop simulation tools that help to reduce costs and development time, while improving safety in the testing phase. Another focal point is the development and experimental validation of localization and state estimation algorithms, whose output is fundamental to maximize the performance of the control system in sport driving conditions. In particular, we propose an alghoritm for the sensor fusion of measurements from inertial and Global Positioning Satellite System units based on an Extended Kalman Filter. Three-dimensional track modeling and trajectory optimization for lap-time minimization are also part of the research. Specifically, we compare the results obtained with the employment of two models of the vehicle dynamics in the three-dimensional space with different levels of accuracy and complexity.
Le competizioni di guida autonoma in circuito offrono un’ottima opportunità per stimolare e sostenere lo sviluppo e la sperimentazione nel campo della guida autonoma, consentendo di spingere i limiti della tecnologia attraverso la competizione e, al contempo, di rimanere all’interno di un ambiente controllato e sicuro. Negli ultimi anni, la popolarità delle competizioni di guida autonoma su circuito, come la Indy Autonomous Challenge, la Roborace e la Formula Driverless, è cresciuta notevolmente. Questo scenario ha portato alla nascita di nuove sfide tecnologiche che richiedono lo sviluppo di sistemi di controllo avanzati e particolarmente performanti per veicoli sportivi e da corsa ed ha incentivato lo studio di soluzioni innovative in particolare sul tema del controllo della dinamica dei veicoli ai limiti di attrito. In questa tesi, ci concentriamo sullo sviluppo e la validazione sperimentale di sistemi di controllo in retroazione per veicoli ad alte prestazioni in condizioni di guida sportiva, includendo tecniche di controllo fortemente basate su modelli dinamici come il Model Predictive Control. Il lavoro si focalizza sull’architettura tradizionale dei veicoli con sterzo, freno e acceleratore come ingressi di controllo, con una estensiva validazione sperimentale ad alta velocità su una macchina da corsa della serie IndyLights. Inoltre, vengono esplorate le potenzialità di architetture alternative utilizzate talvolta nel campo dei veicoli ad alte prestazioni. Ad esempio, ci si concentra sulla tecnica del Torque Vectoring (per veicoli elettrici ed ibridi capaci di controllare la coppia trasmessa alle singole ruote) e la tecnologia delle ruote posteriori sterzanti. Nonostante queste applicazioni siano affrontate solo in simulazione, lo studio si predispone per l’implementazione in tempo reale degli algoritmi di controllo tenendo conto delle limitazioni di potenza di calcolo che caratterizzano le centraline di controllo veicolo. Inoltre, questo lavoro affronta alcune tematiche complementari all’attività di sviluppo delle strategie di controllo della dinamica dei veicoli. Fra queste, in particolare, trattiamo la progettazione di strumenti di simulazione hardware-in-the-loop, utili a ridurre i costi e i tempi di sviluppo e, allo stesso tempo, a garantire la sicurezza nelle fasi di sperimentazione. Un ulteriore punto focale del presente studio riguarda lo sviluppo e la validazione sperimentale di algoritmi di localizzazione e stima dello stato, il cui prodotto è fondamentale per massimizzare le prestazioni del sistema di controllo in condizioni di guida sportiva. In particolare, viene proposto un algoritmo di fusione delle misure inerziali con quelle provenienti da un sistema satellitare globale di navigazione tramite un Extended Kalman Filter. Infine, viene studiato il problema della modellazione tridimensionale dei circuiti e della generazione di traiettorie al fine di minimizzare il tempo sul giro. In particolare, vengono confrontati i risultati ottenuti con due modelli della dinamica veicolo nello spazio tridimensionale di complessità e accuratezza diversa.
Advanced Vehicle Dynamics Control for High-Performance Self-Driving Cars
ALBERTO, LUCCHINI
2023
Abstract
Autonomous racing represents the ideal scenario to boost autonomous driving development and testing, as it pushes the limits of technology through competition, while remaining confined to a controlled and regulated environment. In recent years, self-driving competitions such as Indy Autonomous Challenge, Roborace and Formula Driverless are becoming popular. In this context, several challenges arise, requiring more advanced and performant control systems for sports and racecars and resulting in novel solutions for vehicle dynamics control at the limits of handling. In this thesis, we focus on the development and experimental validation of feedback control systems for high-performance vehicles in sport driving conditions, including strongly model-based control techniques such as Model Predictive Control. The research is based on the traditional architecture with steering and throttle/brake control inputs, with an extended high-speed experimental validation on a full-scale racecar. Besides that, we explore the potentialities of alternative architectures that are common in the world of high-performance vehicles. We focus on torque vectoring (for electric and hybrid vehicles with multiple electric motors) and four-wheel-steering. Despite these latter applications are limited to the simulation environment, attention is placed on the limitation of computational complexity for potential real-time implementation on low-power hardware. In the development of the control strategies, it is fundamental to deal with several side aspects of vehicle dynamics control. In particular, we cover the design of hardware-in-the-loop simulation tools that help to reduce costs and development time, while improving safety in the testing phase. Another focal point is the development and experimental validation of localization and state estimation algorithms, whose output is fundamental to maximize the performance of the control system in sport driving conditions. In particular, we propose an alghoritm for the sensor fusion of measurements from inertial and Global Positioning Satellite System units based on an Extended Kalman Filter. Three-dimensional track modeling and trajectory optimization for lap-time minimization are also part of the research. Specifically, we compare the results obtained with the employment of two models of the vehicle dynamics in the three-dimensional space with different levels of accuracy and complexity.File | Dimensione | Formato | |
---|---|---|---|
thesis.pdf
accesso solo da BNCF e BNCR
Dimensione
20.23 MB
Formato
Adobe PDF
|
20.23 MB | Adobe PDF |
I documenti in UNITESI sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/20.500.14242/297373
URN:NBN:IT:POLIMI-297373