IEEE CEC 2007 Car Racing Competition

Julian Togelius and Simon M. Lucas

In association with the Congress on Evolutionary Computation 2007. There is a prize of 500 USD to the best competitor, generously sponsored by IEEE CIS (subject to at least five entrants registering for the conference).

Overview
Getting started
Technicalities
Inspiration
League table
Rules
Submitting your controller

Last update: September 19, 2007

League table

20.3Ho Duc Thang and Jon Garibaldi
Controller
representation		Non-stationary fuzzy system plus direct control for waypoint choice
Training
method		?
19.5Tomoharu Nakashima
Controller
representation		Neural networks; direct control combined with internal simulation of opponent strategy for way point choice
Training
method		Genetic algorithm, reinforcement learning and handcoding
19.1Tan Chin Hiong
Controller
representation		Magnetic force field controller
Training
method		Force coefficients optimized by a genetic algorithm, plus hand-coded mechanism for way point choice
19.0Binatix, Inc. (Shay Berant and Udi Barzilai)
Controller
representation		Direct critic (state-value function based control) represented as feed-forward neural net
Training
method		Temporal difference learning
18.7The ECU Intelligent Systems Class : P. Hingston, J. P. Barbier, M. Fazio, M. Strange, A. Azemia, L. Ross
Controller
representation		Modular neural / fuzzy hybrid
Training
method		Genetic algorithm + hardcoded domain knowledge
16.9Bob MacCallum
Controller
representation		Plain old Java, direct control
Training
method		Java code evolved with PerlGP (genetic programming) - no other vehicles present (thanks to Imperial College HPC for 90+ CPU days).
16.5Thomas Haferlach
Controller
representation		Modular CTRNN network, direct control
Training
method		Modified CoSyNE algorithm
15.4Aravind Gowrisankar
Controller
representation		Several neural networks with evolved topology, direct control
Training
method		NEAT
15.2Pete Burrow
Controller
representation		Hierarchical modular controller based on two neural networks, direct control
Training
method		Evolution strategy plus hand-coded decision mechanism
14.0Matt Simmerson
Controller
representation		Neural network with evolved topology, direct control
Training
method		NEAT
12.6Julian Togelius
Controller
representation		9 Perceptrons; state-value function based control
Training
method		Temporal difference learning (TD(0))
12.4Julian Togelius
Controller
representation		Neural net (MLP), direct control
Training
method		Evolution strategy
11.2Julian Togelius
Controller
representation		Expression tree (GP), direct control
Training
method		Evolution strategy
The above submissions are ranked in order of competition score. However, the controller with the highest competition score is no necessarilythe winner of the competition. The winner of the competition will be the winner of a tournament between the four highest-ranked controllers, and will be announced at the competition session at CEC.

Remember: the submission deadline is September 20, 2007.

The controller below arrived after the submission deadline, and so was not included in the official competition. However, it is scored here and provided for comparison purposes.
18.5Frank Fuentes
Controller
representation		Straight Java, hand-coded switches and variables
Training
method		None!