This page links to movies, images and other information about my Ph.D. dissertation.
Documents
DissertationJournal Articles
- The Simulated Evolution of Robot Perception. The dissertation.
Conference Papers
- Evolving Visual Sonar: Depth From Monocular Images Aimed at a computer vision audience, this article is to appear in Pattern Recognition Letters.
Proposal
- Genetic Programming For Real World Robot Vision A six page paper aimed at a Robotics audience. Presented at IROS 2002, the International Conference on Intelligent Robots and Systems.
- Genetic Programming for Robot Vision A ten page paper aimed at an Artificial Life audience. Assumes familiarity with Genetic Programming. Presented at SAB 2002, the Seventh International Conference on the Simulation of Adaptive Behaviour.
- Visual Obstacle Avoidance Using Genetic Programming: First Results This early paper is about the expanded representation experiments only. It was presented at GECCO-2001, the Genetic and Evolutionary Computation Conference.
- Breaking Out of the Black Box: A New Approach to Robot Perception. This is a ten page version of my thesis proposal. It was presented at Mobile Robots XIII in 1998.
- Breaking Out of the Black Box: A New Approach to Robot Perception. The thesis proposal (Abstract).
Ground TruthOberon
- A handy java application that allows you to tell Oberon what the correct interpretation of the images are, so it can learn. Available as a windows or Linux executable, or as a Java .jar file. To use the .jar file, you'll need the Java 2 Platform v. 1.4.2, or on a Mac, Java 1.4.2.
- This source archive contains all the code for evolution, as well as code to analyze and simplify the evolved individuals, to run them on new images, and the navigation code that turns them into a direction to travel. It's all in C++, except for the simplifier, which is in LISP.
Focused RepresentationExpanded Representation
- Movies of the best-of-experiment individuals from the three experiments put through their paces on both the training and very similar test data.
Controlling the Robot
- Movies of the best-of-experiment individual put through its paces on both the training and very similar test data.
- Movies of the evolved individuals used to control the robot. This is what the robot sees, with the output of the evolved individuals marked on every frame. An annotated version, with various markings showing what the navigation algorithm is thinking, is also available.