2003 Research Fair Archive - Computer Science Abstracts
Autonomous Robotic Navigation
Several mobile robotic platforms were constructed and used to evaluate the feasibility of implementing exploration and mapping algorithms on embedded processors. Processor constraints included a limited ability to perform accurate calculations, limited memory, and limited support for data structures. An object-oriented software design approach was used to interface multiple sensors to mapping, communication, and motor control modules. Sensors included magnetic compass, sonar ranging and impact collision detection. Design solutions were demonstrated which implemented exploration, mapping and communication protocols within the processor constraints. Techniques for mapping 10 x 10 ft areas to 6 inch precision were developed which could be progressively refined as a mobile robot explored its environment. A wireless token-ring protocol was developed for transmitting stored maps. While mapping algorithms recorded good static positional accuracy, an attempt to produce reliable positional translation independent of external reference points led to the conclusion that dead-reckoning navigation is inadequate for application to autonomous mobile robotics. Over moderate transit distances small errors were destructively cumulative. Two sources of cumulative path error were identified: electromechanical component variability and path calculation errors due to integer precision math. Stored map path control, when decoupled from environmental landmarks, did not provide enough information to provide accurate navigation.