Nowadays, the automotive industry is focused on the challenges posed by Autonomous Vehicles. Among them, the toughest regards the acceptance of the proposed driving logic from the final users. To avoid a possible reject from the passengers, it is important that the driver logic resembles as much as possible the human driver behaviour. This is why many OEMs devoted resources to the understanding and modelling of the human driver. In principle, a human driver model could be formulated by merging various manoeuvre-specific models. In practice, one of the drawbacks of this approach is that the model calibration could be a rather challenging task. This is why it would be better to formulate a comprehensive, unified model. Kolekar et al. were the first authors to create a unified framework for the modelling of the human driver. The model relies on the definition of the Driver's Risk Field (DRF), that represents the human driver's belief about the probability of an event to occur. In this thesis, an effort is made to achieve two milestones. The first is related to the development of a perception model to be integrated in the DRF. The perception model should reproduce how different drivers perceive the external environment and the other road users. The second aims at assessing the possible acceptance from car passengers of the newly proposed driver model. To do so, an experimental campaign with 31 volunteers was performed using a dynamic driving simulator. An ad-hoc experimental protocol to obtain both quantitative (EEG, ECG, SPR and eye-tracker) and qualitative (surveys) answers was developed. The reactions obtained when testing various driver model versions were compared with those obtained using a real human driver.
Al giorno d'oggi, l'industria automobilistica è focalizzata sulle sfide poste dai veicoli a guida autonoma. Tra queste, la più probante è legata all'accettazione delle logiche di guida da parte degli utenti finali. Per evitare un possibile rifiuto da parte dei passeggeri, è importante che la logica di guida si il più vicino possibile al comportamento di un pilota umano. Questo è il motivo per cui molti OEM hanno dedicato risorse al capire e modellare il guidatore umano. In principio, un modello di pilota può essere formulato unendo più sotto-modelli tra loro (modelli che si focalizzano su una manovra specifica). In pratica, però, uno dei punti deboli di questo approccio è legato alla fase di calibrazione. Perciò, sarebbe meglio formulare un modello unificato (e non frammentato). Kolekar et al. sono stati tra i primi a seguire questa strada per la creazione di un modello di pilota. Questo modello è basato sulla idea del Driver's Risk Field (DRF), che rappresenta il parere del guidatore sull'accadere o meno di un certo evento. In questa tesi si è cercato di raggiungere due obiettivi. Il primo riguarda lo sviluppo di un modello di percezione da integrare nel DRF. Questo modello riproduce il modo in cui diversi guidatori percepiscono l'ambiente e gli altri utenti della strada. Il secondo obiettivo mira a capire se il modello sviluppato sia percepito positivamente dai passeggeri. Per fare ciò, sono stati reclutati 31 volontari per una campagna sperimentale usando un simulatore di guida. E' stato sviluppato un protocollo sperimentale per ottenere risposte sia qualitative (questionari) che quantitative (EEG, ECG, SPR ed eye-tracker). Le reazioni dei volontari ottenute testando varie versioni del modello di pilota sono state confrontate con quelle ottenute usando un guidatore umano.
An innovative human-like driver model
Cioffi, Antonio
2023/2024
Abstract
Nowadays, the automotive industry is focused on the challenges posed by Autonomous Vehicles. Among them, the toughest regards the acceptance of the proposed driving logic from the final users. To avoid a possible reject from the passengers, it is important that the driver logic resembles as much as possible the human driver behaviour. This is why many OEMs devoted resources to the understanding and modelling of the human driver. In principle, a human driver model could be formulated by merging various manoeuvre-specific models. In practice, one of the drawbacks of this approach is that the model calibration could be a rather challenging task. This is why it would be better to formulate a comprehensive, unified model. Kolekar et al. were the first authors to create a unified framework for the modelling of the human driver. The model relies on the definition of the Driver's Risk Field (DRF), that represents the human driver's belief about the probability of an event to occur. In this thesis, an effort is made to achieve two milestones. The first is related to the development of a perception model to be integrated in the DRF. The perception model should reproduce how different drivers perceive the external environment and the other road users. The second aims at assessing the possible acceptance from car passengers of the newly proposed driver model. To do so, an experimental campaign with 31 volunteers was performed using a dynamic driving simulator. An ad-hoc experimental protocol to obtain both quantitative (EEG, ECG, SPR and eye-tracker) and qualitative (surveys) answers was developed. The reactions obtained when testing various driver model versions were compared with those obtained using a real human driver.File | Dimensione | Formato | |
---|---|---|---|
Thesis.pdf
accessibile in internet solo dagli utenti autorizzati
Descrizione: PhD Thesis
Dimensione
78.81 MB
Formato
Adobe PDF
|
78.81 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/10589/220533