Automation is one of the most important features of the future industry, with concepts like 5G, machine learning and industry 4.0 becoming more realistic. The automotive industry is a key sector in this field, leading this new trend towards automation by carrying out an immense amount of research an using creative and cutting-edge solutions to be one of the first sectors of the industry to implement autonomous and human-like features in their products. On one hand the rising concern about the environmental impact of cars and on the other hand the issue of traffic and congestion in large cities, has convinced us that the number of cars must be reduced. These issues coupled with the attempt to make safer cars and to reduce the number of accidents have led to one solution, autonomous vehicles. Autonomous driving reduces the number of needed cars and through more efficient driving style, it reduces pollution and congestion, also with crash avoidance capabilities, autonomous vehicles are much safer and are immune to human error. The first step towards autonomous driving is studying the driver behavior and the mechanism through which the human driver perceives the driving environment, interprets it, and then makes decisions and takes actions. The study of driver behavior has led to defining various driver models, these mathematical models are the core cognitive foundation of virtual drivers, implementing these models in intelligent virtual agents leads to an artificial agent with capabilities to control a vehicle. The focus of this thesis is on investigating the behavior of the virtual driver in the IPG CarMaker software. The virtual driver in this software can drive a vehicle along a defined path and respect limitations it encounters which are either defined by the user or by the driving environment. There are various parameters and settings available in the driver module of CarMaker, changing the value of these parameters will consequently modify the behavior of the virtual driver. In this study, by defining various sets of parameters 29 virtual drivers with different behaviors were defined. These virtual drivers then drove around a path in CarMaker, designed to replicate a stretch of urban road. Their behavior was assessed using the output signals of speed, steering, gas and brake pedal and deviation from the path. To have a base for comparison, the virtual drivers were compared with 26 real drivers who had previously driven around the same track in the driving simulator in a former research. The two groups’ driving behavior has been compared by analyzing the signals coming from each driver. The evaluation of their behavior was based on statistical analysis and clustering methods. By clustering similar drivers first within and then between groups, the effect of each parameter on the virtual driver’s behavior was studied and at the end, settings for which the virtual driver shows a more similar behavior to that of the real drivers were identified.
L'automazione è una delle caratteristiche più importanti del futuro settore industriale, con concetti come 5G, apprendimento automatico e industria 4.0 che diventano più realistici. L'industria automobilistica è un settore chiave in questo ambito, guidando questa nuova tendenza verso l'automazione conducendo un'immensa quantità di ricerca e utilizzando soluzioni creative e all'avanguardia per essere uno dei primi settori dell'industria a implementare funzionalità autonome e simili all'uomo nel loro prodotti. Da un lato la crescente preoccupazione per l'impatto ambientale delle automobili e, dall'altro, la questione del traffico e della congestione nelle grandi città, ci ha convinto che il numero delle auto deve essere ridotto. Questi problemi, uniti al tentativo di rendere le auto più sicure e di ridurre il numero di incidenti, hanno portato a un'unica soluzione, i veicoli autonomi. La guida autonoma riduce il numero di auto necessarie e, attraverso uno stile di guida più efficiente, riduce l'inquinamento e la congestione, anche con capacità di prevenzione degli incidenti, i veicoli autonomi sono molto più sicuri e sono immuni all'errore umano. Il primo passo verso la guida autonoma è studiare il comportamento del guidatore e il meccanismo attraverso il quale il guidatore umano percepisce l'ambiente di guida, lo interpreta, quindi prende decisioni e intraprende azioni. Lo studio del comportamento del guidatore ha portato alla definizione di vari modelli di guidatore, questi modelli matematici sono il fondamento cognitivo di base dei guidatori virtuali, l'implementazione di questi modelli in agenti virtuali intelligenti porta a un agente artificiale con capacità di controllare un veicolo. Il focus di questa tesi è indagare il comportamento del guidatore virtuale nel software IPG CarMaker. Il guidatore virtuale in questo software può guidare un veicolo lungo un percorso definito e rispettare i limiti che incontra che sono definiti dall'utente o dall'ambiente di guida. Ci sono vari parametri e impostazioni disponibili nel modulo driver di CarMaker, la modifica del valore di questi parametri modificherà di conseguenza il comportamento del driver virtuale. In questo studio, definendo vari set di parametri, sono stati definiti 29 driver virtuali con comportamenti diversi. Questi piloti virtuali hanno quindi guidato su un percorso in CarMaker, progettata per replicare un tratto di strada urbana. Il loro comportamento è stato valutato utilizzando i segnali di velocità, sterzo, pedale dell'acceleratore e del freno e deviazione dal percorso. Per avere una base di confronto, i piloti virtuali sono stati confrontati con 26 piloti reali che avevano precedentemente guidato sullo stesso percorso con un simulatore di guida in una precedente ricerca. Il comportamento di guida dei due gruppi è stato confrontato analizzando i segnali provenienti da ciascun guidatore. La valutazione del loro comportamento si è basata su analisi statistiche e metodi di clustering. Raggruppando driver simili prima all'interno e poi tra i gruppi, è stato studiato l'effetto di ogni parametro sul comportamento del conducente virtuale e alla fine sono state identificate le impostazioni per le quali il driver virtuale mostra un comportamento più simile a quello dei driver reali.
Investigating the behavior of intelligent virtual drivers
Yousefi, Mohammad Kia
2019/2020
Abstract
Automation is one of the most important features of the future industry, with concepts like 5G, machine learning and industry 4.0 becoming more realistic. The automotive industry is a key sector in this field, leading this new trend towards automation by carrying out an immense amount of research an using creative and cutting-edge solutions to be one of the first sectors of the industry to implement autonomous and human-like features in their products. On one hand the rising concern about the environmental impact of cars and on the other hand the issue of traffic and congestion in large cities, has convinced us that the number of cars must be reduced. These issues coupled with the attempt to make safer cars and to reduce the number of accidents have led to one solution, autonomous vehicles. Autonomous driving reduces the number of needed cars and through more efficient driving style, it reduces pollution and congestion, also with crash avoidance capabilities, autonomous vehicles are much safer and are immune to human error. The first step towards autonomous driving is studying the driver behavior and the mechanism through which the human driver perceives the driving environment, interprets it, and then makes decisions and takes actions. The study of driver behavior has led to defining various driver models, these mathematical models are the core cognitive foundation of virtual drivers, implementing these models in intelligent virtual agents leads to an artificial agent with capabilities to control a vehicle. The focus of this thesis is on investigating the behavior of the virtual driver in the IPG CarMaker software. The virtual driver in this software can drive a vehicle along a defined path and respect limitations it encounters which are either defined by the user or by the driving environment. There are various parameters and settings available in the driver module of CarMaker, changing the value of these parameters will consequently modify the behavior of the virtual driver. In this study, by defining various sets of parameters 29 virtual drivers with different behaviors were defined. These virtual drivers then drove around a path in CarMaker, designed to replicate a stretch of urban road. Their behavior was assessed using the output signals of speed, steering, gas and brake pedal and deviation from the path. To have a base for comparison, the virtual drivers were compared with 26 real drivers who had previously driven around the same track in the driving simulator in a former research. The two groups’ driving behavior has been compared by analyzing the signals coming from each driver. The evaluation of their behavior was based on statistical analysis and clustering methods. By clustering similar drivers first within and then between groups, the effect of each parameter on the virtual driver’s behavior was studied and at the end, settings for which the virtual driver shows a more similar behavior to that of the real drivers were identified.| File | Dimensione | Formato | |
|---|---|---|---|
|
Mohammad Kia Yousefi.pdf
accessibile in internet solo dagli utenti autorizzati
Dimensione
6.15 MB
Formato
Adobe PDF
|
6.15 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/10589/173693