In this thesis, we studied how physical affordances of the environment, such as traversibility for a mobile robot, can be learned. In particular, we studied how the physical properties of the environment, as acquired from range images obtained from a 3D laser scanner mounted on a mobile robot platform, can specify the traversibility affordance. A physics based simulation environment is used during exploration trials, where the traversibility affordances and the relevant features for each behavior are learned through physical interactions with the environment. The prediction accuracy in perceiving the traversibility affordances of the world, which includes several spherical, cylindrical and box shaped objects, is found to be 94 percent. Furthermore, it is observed that the robot uses only 1.1 percent of extracted features while perceiving the affordances. This in turn saves the time 76.6 percent in scanning and 81percent in feature processing. The robot is later tested in a simulated cluttered environment, surrounded by walls. It is able to successfully traverse in the environment, by selecting its behaviors based on the affordances provided, and performing them. The robot was able to avoid from the box shaped objects, and push-roll the spherical ones without making any object detection. In the last set of experiments, the trained affordance-based behavior selection scheme is partially veried in the real world with the Kurt3D robot.
| Identifer | oai:union.ndltd.org:METU/oai:etd.lib.metu.edu.tr:http://etd.lib.metu.edu.tr/upload/12607603/index.pdf |
| Date | 01 September 2006 |
| Creators | Ugur, Emre |
| Contributors | Sahin, Erol |
| Publisher | METU |
| Source Sets | Middle East Technical Univ. |
| Language | English |
| Detected Language | English |
| Type | M.S. Thesis |
| Format | text/pdf |
| Rights | To liberate the content for public access |
Page generated in 0.0021 seconds