This thesis investigates the feasibility of a feedback linearization algorithm based on differential-drive kinematics when applied to a skid-steering robot model. Using Dymola simulations, I compared closed-loop systems composed of the robot kinematic model and the feedback linearization algorithm—first designed for skid-steering robots and subsequently for differential-drive robots, whose models do not depend on slip, an intrinsically variable difficult to estimate. Using parametric identification on data from four scenarios (DDV/SSV × rigid/deformable), I evaluated five aggregation strategies and ranked the candidates using a multi-objective cost function. Key finding: Models trained on deformable terrain conditions generalize better than models from ostensibly more accessible terrain conditions, regardless of controller type. The median frequency-weighted model trained on deformable terrain consistently achieved the lowest average cost across various scenarios. The generalization of the closed-loop system identified from the feedback linearization algorithm based on skid-steering kinematics maintains a slight advantage on rigid terrain. Feedback linearization based on differential-drive kinematics applied to skid-steering robot control is feasible, offering a simplicity-robustness trade-off suitable for field deployment. The core contribution is a systematic cross-scenario evaluation framework for assessing model generalization. Using parametric system identification on closed-loop data from four scenarios (DDV/SSV × rigid/deformable), I evaluated five aggregation strategies and ranked candidates using a multi-objective cost function balancing time-domain fit and frequency-domain consistency. Key finding: Models trained on challenging conditions (deformable terrain) generalize better than models trained on easier conditions, regardless of controller type. The frequency-median model trained on DDV deformable data achieved the lowest average cross-scenario cost (0.528), successfully extrapolating to twice the training velocity and different controller formulations. SSV retained a slight advantage on rigid terrain (0.531 vs 0.537). Simplified DDV-based feedback linearization is feasible for skid-steering robots, offering a compelling simplicity-robustness trade-off for field deployment.
Questa tesi indaga l'ammissibilità di un algoritmo di feedback linearization basato sulla cinematica di un differential-drive vehicle, quando applicato al modello di un robot con una cinematica skid-steering. Utilizzando simulazioni Dymola, ho confrontato i sistemi in anello chiuso composti dal modello cinematico del robot e l'algoritmo di feedback linearization, dapprima pensato per un robot skid-steering e, successivamente, per un differential-drive, il cui modello non dipende dallo scivolamento che è variabile intrinsecamente difficile da stimare. Usando l'identificazione parametrica su dati da quattro scenari (DDV/SSV × rigido/deformabile), ho valutato cinque strategie di aggregazione e classificato i candidati con una funzione di costo multi-obiettivo. Risultato chiave: I modelli addestrati su condizioni di terreno deformabile si generalizzano meglio dei modelli da condizioni apparentemente accessibili, indipendentemente dal tipo di controllore. Il modello a mediana di frequenze addestrato su terreno deformabile ha raggiunto il costo medio fra i vari scenari più basso, in maniera consistente. La generalizzazione del sistema in anello chiuso identificato a partire dall'algoritmo di feedback linearization basato sulla cinematica di un robot skid-steering mantiene un leggero vantaggio su terreno rigido. La feedback linearization basata su un differential drive vehicle per la cinematica di un robot skid-steering è fattibile, offrendo un compromesso semplicità-robustezza per il deployment sul campo.
Analysis of a linearized skid-steering model
MARTENA, GIULIO MARIO
2024/2025
Abstract
This thesis investigates the feasibility of a feedback linearization algorithm based on differential-drive kinematics when applied to a skid-steering robot model. Using Dymola simulations, I compared closed-loop systems composed of the robot kinematic model and the feedback linearization algorithm—first designed for skid-steering robots and subsequently for differential-drive robots, whose models do not depend on slip, an intrinsically variable difficult to estimate. Using parametric identification on data from four scenarios (DDV/SSV × rigid/deformable), I evaluated five aggregation strategies and ranked the candidates using a multi-objective cost function. Key finding: Models trained on deformable terrain conditions generalize better than models from ostensibly more accessible terrain conditions, regardless of controller type. The median frequency-weighted model trained on deformable terrain consistently achieved the lowest average cost across various scenarios. The generalization of the closed-loop system identified from the feedback linearization algorithm based on skid-steering kinematics maintains a slight advantage on rigid terrain. Feedback linearization based on differential-drive kinematics applied to skid-steering robot control is feasible, offering a simplicity-robustness trade-off suitable for field deployment. The core contribution is a systematic cross-scenario evaluation framework for assessing model generalization. Using parametric system identification on closed-loop data from four scenarios (DDV/SSV × rigid/deformable), I evaluated five aggregation strategies and ranked candidates using a multi-objective cost function balancing time-domain fit and frequency-domain consistency. Key finding: Models trained on challenging conditions (deformable terrain) generalize better than models trained on easier conditions, regardless of controller type. The frequency-median model trained on DDV deformable data achieved the lowest average cross-scenario cost (0.528), successfully extrapolating to twice the training velocity and different controller formulations. SSV retained a slight advantage on rigid terrain (0.531 vs 0.537). Simplified DDV-based feedback linearization is feasible for skid-steering robots, offering a compelling simplicity-robustness trade-off for field deployment.| File | Dimensione | Formato | |
|---|---|---|---|
|
Thesis.pdf
accessibile in internet per tutti
Dimensione
2.36 MB
Formato
Adobe PDF
|
2.36 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/10589/246782