Inhalt

•  
   Introduction
•  
  Background
•  
   (Multi-)Agent Systems
•  
   Single Agent Reinforcement Learning
•  
   Growing Neural Gas for Vector Quantization
•  
  Related Work
•  
   Shepherding Approaches
•  
   Discussion of Shepherding Tasks and Approaches
•  
   Approximations for Reinforcement Learning
•  
   Discussion of Approximation Approaches
•  
   The Shepherding Task
•  
   Motivation
•  
   Biological Background
•  
   Description of the Shepherding Task
•  
   Modeling the Shepherding Task as Multiagent System
•  
   Sheep Behavior
•  
   Complexity of the Shepherding Task
•  
   Conclusion
•  
   Single Agent Shepherding
•  
   Foundations
•  
   A Greedy Shepherding Algorithm
•  
   Analysis of the GCC Algorithm
•  
   Conclusion
•  
   Learning Shepherding Behavior
•  
   Adaptive State Aggregation
•  
   Motivation
•  
   Theoretical Model of State Space Abstraction
•  
   General Approach
•  
   From States to State Regions
•  
   Neighborhood Connections
•  
   Adapting the Approximation
•  
   Refining the Approximation
•  
   Stopping Criteria
•  
   Eligibility Traces for State Regions
•  
   Complete Algorithm
•  
   Analysis
•  
   Conclusion
•  
   Adaptive Function Approximation
•  
   Motivation
•  
   Function Approximation for Reinforcement Learning
•  
   Adjusting the Approximation
•  
   Smoothing the Approximation
•  
   Update Rule
•  
   Complete Algorithm
•  
   Computational Complexity
•  
   Comparison GNG-Q vs. I-GNG-Q
•  
   Conclusion
•  
   Evaluation
•  
  Experimental Results
•  
   Experimental Setup
•  
   Comparison of Base Configurations for GNG-Q and I-GNG-Q
•  
   Evaluation of GNG-Q
•  
   Evaluation of I-GNG-Q
•  
   Comparison to Other Approaches
•  
   Advantages of Adaptive Approximations in Unknown Environments
•  
   Shepherding
•  
   Conclusion
•  
  Conclusion and Future Work
•  
   Conclusions
•  
   Future Work
•  
   Bibliography