In this project thesis, a driving assistant system belonging to the ADAS system class is developed to be applied to a model-scale vehicle. In particular, the model has to be able to execute three mains operations autonomously: 1. Obstacle detection and relative avoidance operations through the development of an algorithm built ad-hoc to manage low-cost and low-performance sensors. The key features of this algorithm have to be simplicity and effectiveness. 2. Cliff detection in order to detect sudden changes in road’s altitude 3. Emergency breaking The system must also be able to communicate the sensor’s data with external devices, using a WiFi module. Using the MQTT protocol, furthermore, the system has to be able to transmit these data to a server in a cloud for storing operations and subsequently use these ones for algorithm improvement. The improvement of the algorithm takes place through the use of a Feed Forward neural network, which after being trained through the data from the sensors, has the aim of completely replacing the deterministic algorithm. This aims to test a higher execution’s velocity of it and therefore reduce vehicle analysis times. For the development of the system, various software were used, including MATLAB for the analysis of the exported data, CubeIDE for programming the RealTime system, AWS for storing online data, Jupyter Notebook for the development of the neural network and Python scripts regarding communication scripts.
In questo elaborato viene sviluppato un sistema di assistenza alla guida appartenente alla classe dei sistemi ADAS da applicare su un modellino in scala di un autoveicolo. Nello specifico tale modello deve essere in grado di eseguire tre principali operazioni in completa autonomia: 1. Identificazione degli ostacoli ed esecuzione automatica della manovra evasiva tramite lo sviluppo di un algoritmo ad-hoc per poter interagire con sistemi a basso costo. Caratteristiche chiave di questo algoritmo devono essere semplicità ed efficacia. 2. Identificazione dei precipizi 3. Frenata d’emergenza Il sistema deve essere in grado poi di comunicare i dati dei sensori con l’esterno grazie all’ausilio di un modulo wifi, e tramite il protocollo MQTT, inviare tali dati ad un server in cloud per la raccolta. Successivamente questi vengono utilizzati al fine di analizzare e migliorare l’algoritmo. Il miglioramento dell’algoritmo avviene tramite l’utilizzo di una rete neurale di tipo Feed Forward, che dopo essere stata addestrata tramite i dati provenienti dai sensori, ha l’obiettivo di sostituire in toto l’algoritmo cosi da testare una maggiore velocità di esecuzione e quindi ridurre i tempi di analisi del veicolo. Per lo sviluppo del sistema diversi software sono stati utilizzati, tra questi MATLAB per l’analisi dei dati esportati, CubeIDE per la programmazione del sistema RealTime, AWS per lo storing dei dati online e Jupyter Notebook per lo sviluppo della rete neurale e degli script Python riguardanti gli script di comunicazione.
Design and development of an obstacle avoidance and cliff detection system through a model-scale vehicle
SPAGNUOLO, PARIDE
2021/2022
Abstract
In this project thesis, a driving assistant system belonging to the ADAS system class is developed to be applied to a model-scale vehicle. In particular, the model has to be able to execute three mains operations autonomously: 1. Obstacle detection and relative avoidance operations through the development of an algorithm built ad-hoc to manage low-cost and low-performance sensors. The key features of this algorithm have to be simplicity and effectiveness. 2. Cliff detection in order to detect sudden changes in road’s altitude 3. Emergency breaking The system must also be able to communicate the sensor’s data with external devices, using a WiFi module. Using the MQTT protocol, furthermore, the system has to be able to transmit these data to a server in a cloud for storing operations and subsequently use these ones for algorithm improvement. The improvement of the algorithm takes place through the use of a Feed Forward neural network, which after being trained through the data from the sensors, has the aim of completely replacing the deterministic algorithm. This aims to test a higher execution’s velocity of it and therefore reduce vehicle analysis times. For the development of the system, various software were used, including MATLAB for the analysis of the exported data, CubeIDE for programming the RealTime system, AWS for storing online data, Jupyter Notebook for the development of the neural network and Python scripts regarding communication scripts.| File | Dimensione | Formato | |
|---|---|---|---|
|
2022_12_Spagnuolo.pdf
non accessibile
Descrizione: testo della tesi
Dimensione
8.49 MB
Formato
Adobe PDF
|
8.49 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/10589/197013