| We present a simulation of an autonomous robot that contains a knowledge-based orienteering module capable of carrying out orienteering tasks. A simulated vision system provides the robot with information about its surroundings. The performance of the vision system is modelled by assigning a probability of recognition to each object dependent on the properties of the objects and their distance from the robot. The range and bearing of objects that are recognized is subject to significant errors. The performance of the vision system is tunable to facilitate a variety of experiments.; It is found that, given the probabilistic vision system, a relatively simple set of rules is sufficient to implement effective and efficient orienteering, and that the resulting system is reasonably robust even when the robot's ability to sense its environment suffers from significant errors. We observe that the difficulty of moving to an object is reasonably independent of the difficulty of visually recognizing the object, that the vision system need concentrate on only one object at a time, and that it is possible to do orienteering with a vision system that only reports objects it is sure it has recognized. (Abstract shortened by UMI.)... |
