|
|
說明 | 141 p |
附註 | Source: Dissertation Abstracts International, Volume: 64-03, Section: B, page: 1403 |
| Director: Kazuhiko Kawamura |
| Thesis (Ph.D.)--Vanderbilt University, 2003 |
| Mobile robots are increasingly used for various purposes. One of the critical functions of mobile robots is to be able to represent points of interest in space and reach them. Maps could provide an easy interface for this purpose by providing a world model. However, both the analysis and the implementation of navigation from the map-based perspective are complex. The behavior-based paradigm presents simple yet robust methods without using a world model. Motor-schema theory is such an example within the behavior-based paradigm. Paths and local attractors are dynamically created by using various potential fields. One problem, however, is despite its ease of implementation and robustness, defining points dynamically within this approach is not easy, if not impossible, without a world model |
| This work analyzes the navigation problem without a world model in a bottom-up fashion. At the core of the implementation lies a concept we call the Egocentric Navigation (ENav). In contrast to map-based approaches that use allocentric representations, ENav relies on two egocentric representations to represent the current and the target states of the world. ENav computes a heading direction based on the comparison of these two representations, and continuously carrying out this comparison while moving along the computed heading takes the robot to the vicinity of the target |
| Compared to map-based methods, ENav presents a reactive implementation for the navigation problem. Implementation is not dependent on the availability of a map; hence navigation does not require explicit localization. Points in space are encoded using egocentric representations. The navigation implementation also does not depend on the availability of range information, absolute heading and odometry. For evaluating the performance of ENav several experiments have been conducted |
| If additional information is available such as absolute bearing or a map it is also shown that such information can be easily integrated with ENav to perform more informed navigation. More challenging navigation issues such as sketch-based navigation and information/experience sharing among robots are also discussed |
| School code: 0242 |
主題 | Engineering, Electronics and Electrical |
| Engineering, Mechanical |
| Computer Science |
| Artificial Intelligence |
| 0544 |
| 0548 |
| 0984 |
| 0800 |
ISBN/ISSN | 049633580X |