| ABSTRACT | 第1-9页 |
| 摘要 | 第9-19页 |
| Chapter 1 Introduction | 第19-25页 |
| ·Motivation and Objectives | 第19-20页 |
| ·Problem statements and research questions | 第20-22页 |
| ·Main scientific contributions | 第22页 |
| ·Summary | 第22-23页 |
| ·Organization of the thesis | 第23页 |
| ·Bibliography | 第23-25页 |
| Chapter 2 Background | 第25-52页 |
| ·Games | 第26页 |
| ·Video games vs. classic games | 第26-27页 |
| ·AI and video games | 第27页 |
| ·Research opportunities in game AI | 第27-30页 |
| ·Test-beds of prey/predator game | 第30-35页 |
| ·Dead-End | 第30-32页 |
| ·Pac-Man | 第32-35页 |
| ·Adaptation | 第35-40页 |
| ·Why is adaptation necessary | 第36-37页 |
| ·Two effects of adaptation | 第37-38页 |
| ·Macro and micro adaptation | 第38页 |
| ·Direct and indirect adaptation | 第38-40页 |
| ·Generating challengeable and satisfactory adaptive game opponents | 第40-41页 |
| ·Player Modeling and Strategy-Based Player Modeling | 第41-42页 |
| ·Analysis of "Player Modeling and Opponent Adaptation" by the MECHANISM approach | 第42-47页 |
| ·The MECHANISM approach | 第42-45页 |
| ·Explaining "Player Modeling and Opponent Adaptation" by the MECHANISM approach | 第45-46页 |
| ·Explaining "Strategy-Based Player Modeling and Opponent Adaptation" by MECHANISM | 第46-47页 |
| ·Summary | 第47-48页 |
| ·Bibliography | 第48-52页 |
| Chapter 3 Opponent Adaptation:generating challengeablegame opponents by CI approaches | 第52-100页 |
| ·Adaptation | 第52-53页 |
| ·Adapt from different aspects | 第53页 |
| ·The evolution of game AI | 第53页 |
| ·Three approaches to accomplish adaptation | 第53-55页 |
| ·Existing knowledge-dependent game AI generation | 第55-59页 |
| ·Finite State Machine | 第55-56页 |
| ·Scripting | 第56-57页 |
| ·Expert System | 第57-58页 |
| ·The pros and cons of knowledge-dependent game AI generation techniques | 第58-59页 |
| ·Proposed knowledge-independent game AI generation with CI | 第59-68页 |
| ·The pros and cons of CI | 第60-62页 |
| ·CI and Automatic Game Design/Development | 第62页 |
| ·Monte Carlo | 第62-63页 |
| ·MCTS | 第63-65页 |
| ·UCT | 第65-67页 |
| ·Prerequisites for applying CI approaches of MCTS and UCT | 第67-68页 |
| ·Applying straight-CI to Dead-End | 第68-78页 |
| ·Defined meta-rules | 第68-69页 |
| ·MCTS control of the NPC "DOG" | 第69-76页 |
| ·UCT control of the NPC "DOG" | 第76-78页 |
| ·Result | 第78页 |
| ·Applying straight-CI to Pac-Man | 第78-82页 |
| ·Defined meta-rules | 第79页 |
| ·MCTS control of the NPC "GHOST" | 第79-80页 |
| ·UCT control of the NPC "GHOST" | 第80-81页 |
| ·Result | 第81-82页 |
| ·Proposed knowledge-independent game AI generation with knowledge-based-CI | 第82-91页 |
| ·The pros and cons of knowledge-based CI | 第82-85页 |
| ·Auto knowledge acquisition | 第85-86页 |
| ·ANN | 第86-89页 |
| ·Optimizing with neuro-evolution | 第89-90页 |
| ·Opponent Adaptation by correlating knowledge-based-CI with SBPM | 第90-91页 |
| ·Applying knowledge-based-CI to Dead-End | 第91-94页 |
| ·Knowledge-based-MCTS control of the NPC "DOG" | 第91-92页 |
| ·Knowledge-based-UCT control of the NPC "DOG" | 第92页 |
| ·Optimizing the neuro-controller | 第92-94页 |
| ·Result | 第94页 |
| ·Applying knowledge-based-CI to Pac-Man | 第94-96页 |
| ·Knowledge-based-MCTS control of the NPC "GHOST" | 第94-95页 |
| ·Knowledge-based-UCT control of the NPC "GHOST" | 第95页 |
| ·Optimizing the neuro-controller from CI-created data with neuro-evolution | 第95-96页 |
| ·Result | 第96页 |
| ·Summary | 第96-97页 |
| ·Bibliography | 第97-100页 |
| Chapter 4 Player Modeling | 第100-152页 |
| ·Player Modeling (PM) | 第100-104页 |
| ·Player Modeling | 第101页 |
| ·Why is Player Modeling necessary | 第101-102页 |
| ·Applications area of Player Modeling | 第102-103页 |
| ·Applicable platforms of Player Modeling | 第103-104页 |
| ·Potential problems incurred by Player Modeling and remedies of it | 第104页 |
| ·Strategy-Based Player Modeling (SBPM) | 第104页 |
| ·What to model | 第104-113页 |
| ·Model from player's preference | 第105-107页 |
| ·Model from player's "preferred style of play" in a storytelling game | 第107-108页 |
| ·Model from player's deviation from the optimal choices | 第108-110页 |
| ·Model from player's satisfaction or entertainment | 第110-112页 |
| ·Model from player's utilization of strategy/Strategy-Based Player Modeling | 第112-113页 |
| ·How to model | 第113-115页 |
| ·Trivial approaches for PM | 第113-114页 |
| ·Machine learning approaches for PM | 第114页 |
| ·Machine learning for SBPM | 第114-115页 |
| ·Supervised learning for SBPM | 第115-124页 |
| ·Supervised learning and its algorithms | 第115-121页 |
| ·Supervised learning for SBPM | 第121-123页 |
| ·Evaluation of the opponent AI | 第123-124页 |
| ·Unsupervised learning for SBPM | 第124-127页 |
| ·Unsupervised learning and its algorithms | 第124-127页 |
| ·Unsupervised learning for SBPM | 第127页 |
| ·Supervised and unsupervised SBPM in Dead-End | 第127-138页 |
| ·Supervised SBPM in Dead-End | 第127-130页 |
| ·Unsupervised SBPM in Dead-End | 第130页 |
| ·SBPM in Dead-End with triple-stategy-players | 第130-136页 |
| ·SBPM in Dead-End with single-stategy-players | 第136-138页 |
| ·Result | 第138页 |
| ·Supervised and unsupervised SBPM in Pac-Man | 第138-148页 |
| ·Supervised SBPM in Pac-Man | 第138-144页 |
| ·Unsupervised SBPM in Pac-Man | 第144页 |
| ·SBPM in Pac-Man with triple-stategy-players | 第144-148页 |
| ·Result | 第148页 |
| ·Summary | 第148页 |
| ·Bibliography | 第148-152页 |
| Chapter 5 Opponent Adaptation:generating satisfactorygame opponents from "DDA by CI" approaches | 第152-174页 |
| ·Entertainment modeling | 第152-153页 |
| ·Qualitative approaches | 第153页 |
| ·Quantitative approaches | 第153页 |
| ·Optimizing Player Satisfaction (OPS) | 第153-154页 |
| ·Implicit approaches | 第153-154页 |
| ·Explicit approaches | 第154页 |
| ·Flow and Gameflow with OPS | 第154-157页 |
| ·Flow | 第154-156页 |
| ·GameFlow | 第156页 |
| ·Flow and Gameflow vs. player's satisfaction | 第156-157页 |
| ·Dynamic Difficulty Adjustment (DDA) | 第157-161页 |
| ·DDA and OPS | 第157页 |
| ·Existing DDA approach | 第157-158页 |
| ·Three different DDAs | 第158-159页 |
| ·Proposed "DDA by CI" | 第159-160页 |
| ·Game Level Design with DDA | 第160-161页 |
| ·Proposed "DDA by time-constrained-CI" | 第161-166页 |
| ·The pros and cons of "DDA by time-constrained-CI" | 第161-162页 |
| ·"DDA by time-constrained-MCTS" | 第162-163页 |
| ·"DDA by time-constrained-UCT" | 第163页 |
| ·"DDA by time-constrained-CI" in Dead-End | 第163-164页 |
| ·"DDA by time-constrained-CI" in Pac-Man | 第164-166页 |
| ·Result | 第166页 |
| ·Proposed "DDA by knowledge-based-time-constrained-CI" | 第166-171页 |
| ·The pros and cons of "DDA by knowledge-based-time-constrained-CI" | 第167页 |
| ·"DDA by knowledge-based-time-constrained-MCTS" | 第167页 |
| ·"DDA by knowledge-based-time-constrained-UCT" | 第167页 |
| ·"DDA by knowledge-based-time-constrained-CI" in Dead-End | 第167-170页 |
| ·"DDA by knowledge-based-time-constrained-CI" in Pac-Man | 第170-171页 |
| ·Result | 第171页 |
| ·Summary | 第171-172页 |
| ·Bibliography | 第172-174页 |
| Chapter 6 Conclusions | 第174-184页 |
| ·Answer to research questions | 第174-176页 |
| ·CI used for NPC control in order to create challengeable opponent | 第174-175页 |
| ·Knowledge-based-CI used for NPC control in order to create challengeable opponent | 第175页 |
| ·SBPM by supervised and unsupervised learning | 第175-176页 |
| ·"DDA by CI" approaches used for generating satisfactory game opponents | 第176页 |
| ·Answers to problem statements | 第176-177页 |
| ·Limitations | 第177-178页 |
| ·Applying the MCTS and UCT | 第177-178页 |
| ·Strategy-Based Player Modeling | 第178页 |
| ·Extensibility and future work | 第178-180页 |
| ·Applying Dynamic Tree Search | 第178页 |
| ·Applying affective computing | 第178-179页 |
| ·Further abstraction for Strategy-Based Player Modeling | 第179页 |
| ·Players'satisfaction test | 第179-180页 |
| ·Summary | 第180-181页 |
| ·Bibliography | 第181-184页 |
| Appendices | 第184-189页 |
| Acknowledgements | 第189-190页 |
| List of Published papers | 第190-192页 |
| 详细摘要 | 第192-209页 |
