In the last 10 years important companies as Tesla, Google and Audi have put much effort in the autonomous driving research; in fact, between 2015 and 2017 the estimated investment in autonomous vehicle technology, as determined by the Brookings Institution, is in the order of 80 billions of dollars. Autonomous driving is expected to have an enormous impact on the mobility and transport, motivated by the increased safety, and economic convenience as well as the reduction of accidents and pollution. In particular, studies have found that over 90% of road accidents can be attributed to human error. The elimination of human error through the widespread adoption of autonomous vehicles is thus expected to significantly reduce accidents, with some experts putting the reduction as high as 90%. Moreover, it has been revealed that the adoption of autonomous vehicles in the United States could positively impact the economy by saving 1.3 trillion dollars a year plus 158 billion dollars in fuel costs; productivity would increase of 507 billion dollars, and 488 billion dollars in accident-related expenses would be saved. The aim of this thesis is to test, improve, and combine two pre-existing trajectory planning methods based on Nonlinear Model Predictive Control (NMPC). In particular, the first NMPC-based logic relies on a direct multiple shooting method, and the numerical solution is provided using the ACADO toolkit coupled with the Quadratic Programming solver qpOASES. The second NMPC-based logic relies on a direct single shooting method, and the numerical solution is carried out adopting a metaheuristic Particle Swarm Optimization (PSO) algorithm. After a brief review on available trajectory planning methods, a set of scenarios has been built and tested through the simulation software IPG CarMaker, in order to test the performance of the current logics and improve it. The developed control scheme is composed by the first logic as the main logic, while the second logic, based on the PSO algorithm, has been adopted as an emergency braking logic. The developed control logic combines the advantages of the two logics to ensure a higher robustness. A decision-making function based on a predictive model of risk has been developed, in order to ensure safety and comfort. Finally, the resulting control scheme is tested through a specific set of urban scenarios aiming to analyse the behaviour of the autonomous vehicle (in terms of safety and comfort) in relation to different urban situation and surrounding obstacle motions.
Negli ultimi 10 anni importanti aziende come Tesla, Google e Audi hanno speso molte energie nella ricerca per la guida autonoma; infatti tra il 2015 e il 2017 l'investimento stimato nel campo del veicolo autonomo, secondo quanto stabilito dalla Brookings Institution, è dell'ordine di 80 miliardi di dollari. Si prevede che la guida autonoma avrà un impatto enorme sulla mobilità e sul trasporto, motivato dalla maggiore sicurezza e convenienza economica, nonché dalla riduzione degli incidenti e dell'inquinamento. In particolare, alcuni studi hanno scoperto che oltre il 90 percento degli incidenti stradali può essere attribuito a un errore umano. Si prevede quindi che l'eliminazione dell'errore umano attraverso l'adozione diffusa di veicoli autonomi ridurrà significativamente gli incidenti, per alcuni esperti fino al 90 percento. Inoltre, è stato rivelato che l'adozione di veicoli autonomi negli Stati Uniti potrebbe avere un impatto positivo sull'economia con un risparmio di 1,3 trilioni di dollari all'anno più 158 miliardi di dollari di costi del carburante; la produttività aumenterebbe di 507 miliardi di dollari e si risparmierebbero 488 miliardi di dollari relativi a spese per incidenti. Lo scopo di questa tesi è di testare, migliorare e combinare due metodi di pianificazione della traiettoria preesistenti basati su Nonlinear Model Predictive Control (NMPC). In particolare, la prima logica del tipo NMPC si basa su un metodo diretto di multiple shooting e la soluzione numerica viene ricavata utilizzando il toolkit ACADO abbinato al solutore qpOASES basato su un Quadratic Programming. La seconda logica del tipo NMPC si basa su un metodo diretto di single shooting e la soluzione numerica viene trovata adottando un algoritmo metaheuristico noto come Particle Swarm Optimization (PSO). Dopo una breve rassegna dei metodi di pianificazione della traiettoria disponibili, è stata definita e testata una serie di scenari attraverso il software di simulazione IPG CarMaker, al fine di testare le prestazioni delle logiche attuali e migliorarle. Lo schema di controllo sviluppato è composto dalla prima logica come logica principale, mentre la seconda logica, basata sull'algoritmo PSO, è stata adottata come logica di frenata di emergenza. La logica di controllo sviluppata combina i vantaggi delle due logiche per garantire una maggiore robustezza. Al fine di garantire sicurezza e comfort, è stata sviluppata una funzione decisionale basata su un modello predittivo di rischio. Infine, lo schema di controllo che ne risulta viene testato attraverso una serie specifica di scenari urbani che mirano ad analizzare il comportamento del veicolo autonomo (in termini di sicurezza e comfort) in relazione alle diverse situazioni urbane e ai movimenti degli ostacoli circostanti.
Virtual validation and improvement of MPC control logics based on SQP and PSO methods
LORELLO, MARCO
2018/2019
Abstract
In the last 10 years important companies as Tesla, Google and Audi have put much effort in the autonomous driving research; in fact, between 2015 and 2017 the estimated investment in autonomous vehicle technology, as determined by the Brookings Institution, is in the order of 80 billions of dollars. Autonomous driving is expected to have an enormous impact on the mobility and transport, motivated by the increased safety, and economic convenience as well as the reduction of accidents and pollution. In particular, studies have found that over 90% of road accidents can be attributed to human error. The elimination of human error through the widespread adoption of autonomous vehicles is thus expected to significantly reduce accidents, with some experts putting the reduction as high as 90%. Moreover, it has been revealed that the adoption of autonomous vehicles in the United States could positively impact the economy by saving 1.3 trillion dollars a year plus 158 billion dollars in fuel costs; productivity would increase of 507 billion dollars, and 488 billion dollars in accident-related expenses would be saved. The aim of this thesis is to test, improve, and combine two pre-existing trajectory planning methods based on Nonlinear Model Predictive Control (NMPC). In particular, the first NMPC-based logic relies on a direct multiple shooting method, and the numerical solution is provided using the ACADO toolkit coupled with the Quadratic Programming solver qpOASES. The second NMPC-based logic relies on a direct single shooting method, and the numerical solution is carried out adopting a metaheuristic Particle Swarm Optimization (PSO) algorithm. After a brief review on available trajectory planning methods, a set of scenarios has been built and tested through the simulation software IPG CarMaker, in order to test the performance of the current logics and improve it. The developed control scheme is composed by the first logic as the main logic, while the second logic, based on the PSO algorithm, has been adopted as an emergency braking logic. The developed control logic combines the advantages of the two logics to ensure a higher robustness. A decision-making function based on a predictive model of risk has been developed, in order to ensure safety and comfort. Finally, the resulting control scheme is tested through a specific set of urban scenarios aiming to analyse the behaviour of the autonomous vehicle (in terms of safety and comfort) in relation to different urban situation and surrounding obstacle motions.File | Dimensione | Formato | |
---|---|---|---|
LorelloThesis_consegna_14_5.pdf
non accessibile
Descrizione: testo della tesi
Dimensione
15.9 MB
Formato
Adobe PDF
|
15.9 MB | Adobe PDF | Visualizza/Apri |
LorelloThesis_final_version_21_05.pdf
non accessibile
Descrizione: testo della tesi
Dimensione
15.98 MB
Formato
Adobe PDF
|
15.98 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/10589/154389