文摘
英文文摘
上海交通大学学位论文原创性声明
上海交通大学学位论文版权使用授权书
Abbreviations
Chapter One Introduction
1. Objective
2. Learning in Soccer Domain
3. Approach
Chapter Two A Literature Survey: Machine Learning of Multi Agents for RoboCup Soccer Domain
1. Introduction
2. RoboCup
2.1. Overview
2.2. What's RoboCup?
2.3. Simulated League
3. Paradigms of agent architecture
4. Reinforcement Learning
4.1. Markov Decision Process (MDP)
4.2. Q-learning
5. Multi Agent Systems
5.1. Reasons to use Multiagent Systems
5.2. Multiagent Systems
5.3. Advantages
5.3. MAS in Robotic Soccer
6. Robotic Soccer as a Reinforcement Learning Problem
6.1. Genetics-based machine learning (GBML)
6.2. Genetic programming
6.3. Radial Basis Function Networks (RBFN)
6.4. Minimax-Q learning
6.5. Behavioral diversity
6.6. Layered learning by Neural Nets
6.7. Hexcer by Adversarial RL
6.8. Variable Learning Rate
7. Conclusion
Chapter Three Agent Oriented Paradigm and Implementation of an Environmental Agent for Soccer Server
1. Introduction
2. Definitions
2.1 Example 1
2.2 Example 2
3. Environmental Agent
3.1 Lower-agents
3.2 Upper-agents
3.3. Environmental-agents
3.4. An example
4. An Interpreter
5. Soccer Agents
5.1. Results
6. Conclusions and future work
Chapter Four Machine Learning
1. Supervised Learning
2. Reinforcement Learning
2.1 Single Agent Reinforcement Learning
2.2 Infinite/Multi agents Reinforcement Learning
2.3. Using Function Approximates
3. Example
4. Conclusion
Chapter Five Function Approximation with Radial Basis Function Networks (RBFN)
1. Linear Combination of Radial Basis Vectors
2. Different Error Measures
2.1 Ridge Regression
2.2. Weighted error criteria
2.3. Cross-Validation
3. Online Adaptation of RBF Weight Parameters
4. Fuzzy Reasoning
5. Clustering as Antecedent Induction
5.1 Hard k-means Clustering
5.2. Soft k-means Clustering
5.3. Expectation Maximization Clustering
5.4. Gradient Descent Clustering
Chapter Six Tree Based Representation
1. K-d Tree Assisted Memory Based Learning
1.1. Kernel Regression
2. Radial Basis Function Trees
2.1. Finding maximum Response of a RBF unit within a hyper rectangle
2.2. Partitioning
Chapter Seven Learning to Shoot to Goal
1. Generation of Training through Simulation
2. Forming the Training Set
3. Evaluation of Goal Scoring Probabilities
3.1. Probability of Ball within Goal
3.2. Probability of Interception
4. Implementation and Results
5. Conclusion
Chapter Eight Learning to Intercept the Ball
1. Input Vector and the Generation of Training Cases
2. Evaluation of Interception Probability and Q values.
2.1. Rewards and Penalties
3. Implementation and Results
4. Conclusion
Chapter Nine Learning to Dribble the Ball
1. Introduction
2. Related Work
3. The Learning Process
3.1. Approach
3.2. Preliminary Skills
3.3. Learning to Dribble
3.4. Soccer Server Simulations and the Coach Client
3.5. Radial Basis Functions
4. Results
5. Conclusion and Future Work
Chapter Ten Derived Behaviors
1. Behaviours from Goal Shooting Task
2. Behaviours from Interception Task
3. Conclusion
Conclusion
Bibliography
Publications
Acknowledgements