Implementazione di una pipeline basata su machine learning per un semplice task di visual servoing per un robot mobile

Toniolo, Stefano (2017) Implementazione di una pipeline basata su machine learning per un semplice task di visual servoing per un robot mobile. Bachelor thesis, Scuola Universitaria professionale della Svizzera italiana (SUPSI).

Full text not available from this repository.

Abstract

In the field of machine learning and robotics, IDSIA’s researchers would like to develop a complete pipeline for a simple task which consists in controlling a mobile robot inside an environment. This system will receive as input an image stream and every frame is processed independently to extrapolate controlling information. The information we want to extract is about obstacles in front of the robot which should be avoided. The approach to this problem is "perception-action" which means that the controller binds directly the perception of the robot to his movements. The machine learning phase is developed over two main techniques, the first one based on classification through random forests, and the second with a convolutional neural network, while the controller is "reactive", that means it relies only on information provided at the current time. As a results, we obtained a fully functional system based on a neural network, which can detect, through real time RGB image stream, most of the obstacles in a real world environment and react telling the robot how to avoid them. Furthermore this pipeline is hardware independent, that means it can work with every other system if they provide an image stream.

Item Type: Thesis (Bachelor)
Supervisors: Gambardella, Luca Maria and Giusti, Alessandro
Subjects: Informatica
Divisions: Dipartimento tecnologie innovative > Ingegneria informatica
URI: http://tesi.supsi.ch/id/eprint/1771

Actions (login required)

View Item View Item