Next:
Introduction
Up:
Evolution of Complexity in
Previous:
Conclusions
CHAPTER 3
Coevolving Behavior with Live Creatures
Introduction
Background and Related Work
Coevolution
Too Many Fitness Evaluations
Learning to Play Games
Self-Play or Play against People?
Intelligence on the Web
A space where agents can thrive and evolve
Experimental Model
Tron Light Cycles
Initial Experiment
System Architecture
Tron Agents
Java Applet
Evolving Agents: The Tron Server
Pseudocode of the foreground Tron server
Fitness Function
Novelty Engine
Feedback from Main Population to Novelty Engine
Tunable Parameters
Pseudocode of the Novelty Engine
Results
Win Rate (WR)
Statistical Relative Strength (RS)
Paired Comparisons Analysis
Analysis of Results
Distribution of Players
Are New Generations Better?
Learning
Evolution as Learning
Human Behavior
Measuring Progress in Coevolution
From the Red Queen Effect to Statistical Fitness
New Fitness Measure for the Main Population
Evolving Agents Without Human Intervention: A Control Experiment
Experimental Setup
Results
Tuning up the Novelty Engine
Test Against Humans
The Huge Round-Robin Agent Tournament
Emergent Behaviors
Standard Complexity Measures
Analysis of sample robots
An Advanced Agent
Emergent Behaviors
Spiral inwards
Live and let live
Staircasing
Opening moves
The cutoff maneuver
Topological games
Combined behaviors
Edging
Space Filling
Unforced errors
Behavior or Imagination?
Quantitative Analysis of Behaviors
Differences Between Human and Agent Behaviors
Maze Navigation
Discussion
Adapting to the Real Problem
Evolution as Mixture of Experts
Human-Machine Coevolution
Human Motivations
Pablo Funes
2001-05-08