While driving in an urban scenario, the motion in large of the vehicle can be described knowing the lateral position and the orientation with respect to the centre of the lane, together with the side slip angle and the yaw velocity. Those quantities represent the state variables on which the autonomous driving control loop should be closed, nevertheless, not all of them can be directly measured by sensors and discrete sample rates must be managed. In the context of the TEINVEIN project, this thesis work deals with the vehicle state estimation problem applied to autonomous driving.\\ Three different algorithms have been designed and implemented on an experimental vehicle. The first one consists of a fuzzy logic based on vehicle kinematics to cope with sensor fusion requirements: it allows obtaining reliable results while reducing the computational burden. The other two approaches are based on Kalman filter state estimation algorithms such as the Extended and Unscented Kalman Filters (EKF, UKF respectively). An innovative adaptation method for the noise measurement covariance matrix is presented. These nonlinear estimators allow taking into account different sample rates, in a soft real time implementation, to provide state estimation at the desired frequency. The hdop (horizontal dilution of precision), related to the quality of the GPS signal, is exploited in all the presented estimation algorithms to correct the position in case of GPS signal losses. Comparisons between the proposed approaches have been provided referring to experimental results.
Durante la guida in uno scenario urbano, il moto in grande del veicolo può essere descritto conoscendo la posizione laterale e l’orientazione rispetto al centro corsia, insieme all’angolo di slittamento laterale e alla velocità d’imbardata. Tali quantità rappresentano le variabili di stato su cui deve essere chiuso il circuito di controllo per la guida autonoma, tuttavia, non tutte possono essere misurate direttamente dai sensori e le frequenze di campionamento a tempo discreto devono essere gestite. Nel contesto del progetto TEINVEIN, questo elaborato di tesi tratta il problema della stima dello stato del veicolo applicato alla guida autonoma. Tre differenti algoritmi sono stati progettati ed implementati su un veicolo sperimentale. Il primo consiste in una logica fuzzy basata sulla cinematica del veicolo per far fronte ai requisiti di sensor fusion: questo permette di ottenere risultati affidabili mentre viene ridotto il costo computazionale. Gli altri due approcci sono basati su algoritmi di stima dello stato non lineari quali l’Extended e l’Unscented Kalman Filters (EKF, UKF rispettivamente). Un innovativo metodo adattivo per la matrice di covarianza dei rumori delle misure è presentata. Questi stimatori non lineari permettono di considerare differenti frequenze di campionamento, in una implementazione soft real time, per fornire una stima della stato alla frequenza desiderata. I confronti tra gli approcci proposti sono stati forniti facendo riferimento ai risultati sperimentali.
Sensor fusion and vehicle state estimation for autonomous driving
ERCOLI, GIANLUCA
2018/2019
Abstract
While driving in an urban scenario, the motion in large of the vehicle can be described knowing the lateral position and the orientation with respect to the centre of the lane, together with the side slip angle and the yaw velocity. Those quantities represent the state variables on which the autonomous driving control loop should be closed, nevertheless, not all of them can be directly measured by sensors and discrete sample rates must be managed. In the context of the TEINVEIN project, this thesis work deals with the vehicle state estimation problem applied to autonomous driving.\\ Three different algorithms have been designed and implemented on an experimental vehicle. The first one consists of a fuzzy logic based on vehicle kinematics to cope with sensor fusion requirements: it allows obtaining reliable results while reducing the computational burden. The other two approaches are based on Kalman filter state estimation algorithms such as the Extended and Unscented Kalman Filters (EKF, UKF respectively). An innovative adaptation method for the noise measurement covariance matrix is presented. These nonlinear estimators allow taking into account different sample rates, in a soft real time implementation, to provide state estimation at the desired frequency. The hdop (horizontal dilution of precision), related to the quality of the GPS signal, is exploited in all the presented estimation algorithms to correct the position in case of GPS signal losses. Comparisons between the proposed approaches have been provided referring to experimental results.| File | Dimensione | Formato | |
|---|---|---|---|
|
2019_10_Ercoli.pdf
non accessibile
Descrizione: Testo della tesi
Dimensione
18.76 MB
Formato
Adobe PDF
|
18.76 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/10589/149746