This work deals with the implementation of a general procedure for the optimal control of an autonomous vehicle. The specific task of the vehicle is to completely clean the floor of a supermarket, an indoor environment characterised by the presence of many obstacles. These obstacles have to be approached tangentially in order to avoid collisions and possible damage of the products. The procedure consists of a first offline part and of a second online one. In the offline phase, given the map of the plant, an optimal trajectory is computed, by means of two levels of computation. In the first level, the space is decomposed as a grid and then the first reference path is computed as a Hamiltonian cycle which minimizes the number of turns, through a Combinatorial Problem modelled in an AMPL environment. In the second level, since the dynamic nonlinear model of the vehicle is introduced, energy and time are minimized with multiple shooting technique in a CasAdi environment. In the online phase the previously computed path is given as input and the aim becomes to avoid the contact with the obstacles and to optimally control the vehicle. Firstly, an online localization technique is implemented by means of an extended Kalman filter. Secondly, a linear quadratic regulator is used for a first control and then a nonlinear model predictive control is developed in order to minimize the deviation from the optimal path
Questo lavoro si occupa dell’implementazione di una procedura generale per il controllo ottimale di un veicolo autonomo. Il compito specifico del veicolo consiste nel pulire completamente il pavimento di un supermercato, un ambiente indoor caratterizzato dalla presenza di molti ostacoli. Questi ostacoli devono essere avvicinati tangenzialmente per evitare collisioni e possibili danni ai prodotti. La procedura consiste in una prima parte offline ed in una seconda online. Nella fase offline, data la mappa dell’impianto, viene calcolata unatraiettoria ottimale, attraverso due livelli di calcolo. Nel primo livello, lo spazio viene scomposto come una griglia e il primo percorso di riferimento viene quindi calcolato come un ciclo hamiltoniano che minimizza il numero di virate, attraverso un Combinatorial Problem modellato in ambiente AMPL. Nel secondo livello, poiché viene introdotto il modello dinamico non lineare del veicolo, l’energia e il tempo vengono minimizzati con la tecnica del multiple shooting in ambiente CasAdi. Nella fase online viene dato in input il percorso precedentemente calcolato e l’obiettivo diventa quello di evitare il contatto con gli ostacoli e di controllare in modo ottimale il veicolo. In primo luogo, viene implementata una tecnica di localizzazione online mediante un filtro di Kalman esteso. In seguito, viene usato un LQR per un primo controllo e, come ultimo passaggio, viene sviluppato un controllo predittivo a modello non lineare per minimizzare la deviazione dal percorso ottimale.
Optimal trajectory and control of autonomous vehicles for cleaning application
Silvestri, Pietro
2021/2022
Abstract
This work deals with the implementation of a general procedure for the optimal control of an autonomous vehicle. The specific task of the vehicle is to completely clean the floor of a supermarket, an indoor environment characterised by the presence of many obstacles. These obstacles have to be approached tangentially in order to avoid collisions and possible damage of the products. The procedure consists of a first offline part and of a second online one. In the offline phase, given the map of the plant, an optimal trajectory is computed, by means of two levels of computation. In the first level, the space is decomposed as a grid and then the first reference path is computed as a Hamiltonian cycle which minimizes the number of turns, through a Combinatorial Problem modelled in an AMPL environment. In the second level, since the dynamic nonlinear model of the vehicle is introduced, energy and time are minimized with multiple shooting technique in a CasAdi environment. In the online phase the previously computed path is given as input and the aim becomes to avoid the contact with the obstacles and to optimally control the vehicle. Firstly, an online localization technique is implemented by means of an extended Kalman filter. Secondly, a linear quadratic regulator is used for a first control and then a nonlinear model predictive control is developed in order to minimize the deviation from the optimal pathFile | Dimensione | Formato | |
---|---|---|---|
2023_05_Silvestri_executive_summary.pdf
non accessibile
Descrizione: executive summary
Dimensione
781.61 kB
Formato
Adobe PDF
|
781.61 kB | Adobe PDF | Visualizza/Apri |
2023_05_Silvestri_thesis.pdf
non accessibile
Descrizione: thesis
Dimensione
5.45 MB
Formato
Adobe PDF
|
5.45 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/10589/212023