SCS Undergraduate Thesis Topics
|Alex Grubb||Paul Rybski||Autonomous Discovery of Landmark Objects|
The goal of this project is to develop a method by which an autonomous mobile robot can discover and learn representations and locations for immobile, landmark objects within an unknown environment. These landmark objects are large, rigid objects within the environment which remain in relatively fixed locations and can later be used for localization or navigation. Examples of such landmark objects include furniture and large decorations such as paintings.
This project attempts to bridge the gap between the problems of autonomous mapping and object recognition in mobile robots. Current mapping algorithms are able to robustly learn the physical structure of an environment and use it for localization, but are unable to learn about the higher level structure, such as which parts of the structure belong to individual objects. Additionally, existing object recognition methods work very well when given a good set of training images describing the objects to be recognized, but there are not many methods for easily obtaining such training sets. By having a mobile robot autonomously discover landmark objects we would be able to simultaneously learn about the higher level structure of the environment and obtain data which can be used by existing object recognition methods. This allows the robot to automatically learn about critical objects within the environment it has to work in, eliminating the requirement for any pre-programmed obj! ect recognition databases or environment maps.