Next: Localization Up: Related Work and State-of-the-art Previous: Bidding Mechanisms
The mapping problem is regarded as one of the most important problems in the field of autonomous robotics, and it dates back to SRI's famous Shakey robot [54]. A robot operating autonomously needs to answer the three basic questions about mapping and navigation, as posited by Levitt and Lawton [39]:
This would be easy if an a priori map were available, but we are dealing with the scenario of unknown environments. That is, the robot has no knowledge at all about what the environment looks like, where the landmarks, the obstacles, etc are. To be able to answer these questions and, thereby, be able to perform its task, the robot must acquire a model of the environment in which it has to navigate through. Recent research on modeling unknown environments is based on two main approaches: occupancy grid-based (or metric), and topological maps.
Another distinctive and very important feature of mapping approaches is localization. The localization problem can be split in two very different particular problems: local localization and global localization. Local localization, also known as position tracking, aims at compensating odometric errors occurring during robot navigation. On the other hand, global localization is concerned with the problem of finding out where a robot is relative to a map of the whole environment. In this thesis we tackle the problem of global localization. There are two main approaches for solving it: model matching and landmark based localization.
In the rest of this section we will go through all these approaches, starting with the global localization approaches, and then the grid-based and topological mapping ones.