This thesis is broadly concerned with the representation of the environment of a mobile robot and the modelling of its motion. An attempt has been made to address some issues of the laser scan matching for global self-localization and map building. Different methods for the interpretation of sensor information have been investigated. Mobile robots have many applications in transportation, surveillance, health care and mining etc. For a successful navigation, the representation of the environment is crucial. The robot environment interaction is very complex in practice. Many factors contribute to this complexity, such as the electromechanical hardware structure and complex controlling and navigational programming modules. Above all however, it is the environment itself which is usually very complex. The perception model is the most important component of the navigation system of a mobile robot, at the core of which is the representation of the environment. Environment parameters are difficult to model and simplistic models are used in various position estimation techniques. However, for true autonomous navigation, the environment should be represented in a more dense fashion and the interpretation should be straightforward. The robot interacts with its environment using sensors. The sensory information provides clues about the location of the robot but the interpretion of this information is very challenging. Some type of model or a mathematical description of the environment is required for any meaningful interpretation and for making critical navigational decisions when a new observation arrives. The second key component of a navigation system is a motion model. Due to structural and software complexity the behaviour of a robot is rarely repeatable under the same motion commands. This can be attributed to many factors such as slippage, wear and tear of wheels at different rates, floor conditions or obstacle negotiation strategies. This means that motion commands have an associated uncertainty and need statistical treatment. Similarly the processing of raw laser data, although highly desirable, is computationally very expensive and therefore we usually need to make a trade off and extract some features from this data, despite losing some of the information. In this thesis we investigated three core issues of motion modelling, perception (or observation) modelling and scan correlation. Some auxiliary issues have also been addressed, such as the extraction of features from laser data and a broader classification of the environment suitable for certain situations. In regard to environment representation, we used the geometrical form of representation and tried to extract some statistical formulation. This method suggests to capture the environment model in a statistical form before the start of navigation when the map is known. The detailed parametric representation of the environment is obtained along with a proposal for a laser scan matching method based on geometrical line and corner features. The geometrical representation is based on some features extracted from raw laser data. This is considered a compact and easily implementable form, which was one of the objectives of our research, however utilisation of all the sensory information is still desirable and we have also investigated this issue. The models have been tested thoroughly on simulations and with real data in laboratory and office-like indoor environments. Laser scan matching is a technique of position estimation based on matching two laser scans taken at the initial and final positions of the robot. We also presented a method to find out the degree of match between two laser scans. At the end of the thesis, the scan correlation has been used to find the most reliable landmarks in the environment. This approach filters out the nuisance landmarks which increase the size of matrices in Simultaneous Localization and Mapping algorithms. An improved computational efficiency was of primary concern and a main focus of this research. All the methods proposed in this thesis, such as feature extraction, broader classification, parametric formulation, line segment based scan matching and the scan matching for measurement updates address the computational issues in a fundamental way by using an appropriate formulation of the problem.
| Identifer | oai:union.ndltd.org:ADTP/187614 |
| Date | January 2008 |
| Creators | Yaqub, Tahir, Mechanical & Manufacturing Engineering, Faculty of Engineering, UNSW |
| Source Sets | Australiasian Digital Theses Program |
| Language | English |
| Detected Language | English |
| Rights | http://unsworks.unsw.edu.au/copyright, http://unsworks.unsw.edu.au/copyright |
Page generated in 0.0019 seconds