Advanced Robotic System Enhanced with Computer Vision

OData support
Nagy Ákos
Department of Automation and Applied Informatics

Based on recent scientific research regarding manufacturing science, it is presumable that the spread of cyber-physical production systems will bring great industrial progress since the classical robotic systems - working in a carefully designed and highly structured environment without many sensors - are not competent to address the novel problems of robotics. The robots of a modern factory have to be capable of working in a semi-structured environment, collaborating with humans and other machines, and optimizing their own performance by learning from themselves and from other devices as well. This thesis presents a system design of an advanced robotic entity, which is armed with all the necessary tools to tackle these problems. Correspondingly, the entity is capable of perceiving its environment, structuring the gathered knowledge, and acting in order to maximize its performance. These attributes are fulfilling the definition of rational agents which are at the focus of the recent artificial intelligence research.

One of the main features of an intelligent entity is its essential ability to explore its environment. The most efficient way of sensing is processing visual data which is a crucial problem of modern robotics. Traditional systems based on machine vision not only require the extremely careful design of illumination settings but can only be applied for the given application. In contrast, the novel field of computer vision attempts to model human vision and since it learns a general representation of the objects, it not only can be used for various applications, but the expensive and time-consuming illumination design can be omitted as well. Exploiting the recent breakthroughs of computer vision performed in other domains, this thesis enhances the aforementioned intelligent entity with a computer vision module using a deep neural network.

Adaptive assembly is an unsolved problem of manufacturing. In order to assess the system performance, case studies are designed and presented addressing this problem. Thus, this thesis can also offer a strong basis for scientific research on adaptive assembly.


Please sign in to download the files of this thesis.