Technical Report Number
In this thesis, we describe the use of Genetic Programming (GP) to learn obstacle detectors to be used for obstacle avoidance on a mobile robot. The first group of experiments focus on learning visual feature detectors for this task. We provide experimental results across a number of different environments, each with different characteristics, and draw conclusions about the performance of the learned feature detector and the training data used to learn such detectors. We also explore the utility of seeding the initial population with previously evolved individuals and subtrees, and discuss the performance of the resulting individuals. We then include sensory data from a laser range-finder and a camera and discuss the performance of resulting individuals as we use just laser data, just image data, and both in combination.
Marek, Andrew, "Learning Feature Detectors Using Genetic Programming With Multiple Sensors" Report Number: WUCSE-2004-22 (2004). All Computer Science and Engineering Research.
Permanent URL: http://dx.doi.org/10.7936/K7ST7N6V