Publications

Sort by: [year] [type] [author]

Learning Visual Feature Detectors for Obstacle Avoidance using Genetic Programming

Andrew J. Marek, William D. Smart, and Martin C. Martin.
In "Proceedings of the Genetic and Evolutionary Computation Conference (GECCO 2002) Late-Breaking Papers Track", New York, NY, July 2002.

In this paper, we use Genetic Programming (GP) techniques to learn visual feature detectors for a mobile robot navigation task. We provide results from a number of environments, each with different characteristics, and draw conclusions about the performance and nature of the training and testing data. We explore the utility of seeding the initial population with a previously evolved individual, and compare this to previous work, where a hand-coded individual was successfully used as an initial seed. Our experiments exhibited a peculiar bi-modality in final performance. An individual either performed very well or very poorly, with nothing in between. We analyze this phenomenon, and offer suggestions as to its cause, and how to alleviate the problem. We explore the utility of seeding the initial population with a previously evolved individual, and compare this to previous work, where a hand-coded individual wassuccessfully used as an initial seed.

Paper: [PDF]

@inproceedings{gecco02,
  author = {Marek, Andrew J. and Smart, William D. and Martin, Martin C.},
  title = {Learning Visual Feature Detectors for Obstacle Avoidance using Genetic Programming},
  booktitle = {Proceedings of the Genetic and Evolutionary Computation Conference ({GECCO 2002}) Late-Breaking Papers Track},
  address = {New York, NY},
  month = {July},
  year = {2002}
}