An implementation of the Electric Field Approach applied to the simulated RoboCup is presented, together with a demonstration of a learning system. Results are presented from the optimization of the Electric Field parameters in a limited situation, using the learning system. Learning techniques used in contemporary RoboCup research are also described including a brief presentation of their results.
Contents
1 Introduction
1.1 RoboCup
1.1.1 Simulated league
1.1.2 Environmental issues
1.2 Electric Field Approach
1.3 Learning techniques
1.3.1 Reinforcement learning
1.3.2 Q-learning
1.3.3 Hill-climbing
1.4 Contemporary research
1.4.1 Brainstormers
1.4.2 Tsinghuaeolus
1.4.3 CMUnited
1.5 Problem description
1.6 Delimitations
1.7 Method
1.7.1 Literature survey
1.7.2 Experiments
1.8 Thesis outline
2 Implementation
2.1 CRaPI, a RoboCup API
2.2 Yaffa, a RoboCup player
2.3 Our approach
2.3.1 Conceptualization of EFA
2.3.2 Implementation of EFA
2.3.3 Implementation of a learning system
3 The experiment
3.1 Set up
3.2 Results
3.2.1 Training phase
3.2.2 Benchmarks
4 Discussion
4.1 Results
4.1.1 Utilities for training
4.1.2 Trained vs Untrained
i4.1.3 Game results
4.1.4 Reliability
4.2 Problems
4.2.1 Passing
4.2.2 Intersecting
4.2.3 WorldModel
4.2.4 Size of implementation
5 Conclusion
5.1 Future work
6 Acknowledgements
Bibliography
Author: Markus Bergkvist, Tobias Olandersson
Source: Blekinge Institute of Technology
Reference URL 1: Visit Now