Ertugrul Akay | Queen Mary, University of London (original) (raw)
A Computer Science Master student with an interest in General Game AI Research.
less
Uploads
Papers by Ertugrul Akay
Modern Board games are great mediums for game Artificial Intelligence (AI) research since they in... more Modern Board games are great mediums for game Artificial Intelligence (AI) research since they include complex problems to solve within a set of rules and have various interactable components. Tabletop Games Framework is used to digitalize and implement the board game called Battlelore: Second Edition for this purpose. It is the first wargame to have in the framework and an interesting challenge for different game AI agents. This paper summarizes the process of designing and implementing the board game and numerous gameplaying agents like Rule-Based, One Step Look Ahead, Monte Carlo Tree Search (MCTS) and Rolling Horizon Evolutionary (RHEA) algorithms. After the implementation of the core game with some minor constraints in main features and components, the performance of implemented agents is observed through a Round Robin Tournament. The findings from this experiment are in correlation with the hypothesis for some agents, except for one outperforming (MCTS) and one underperforming agent (Rule-Based). Further work is possible to have betterperforming agents, or some of the game parameters can be tweaked to use the game in different research areas.
In this paper, the concept of using agent behaviors in the 3D environment of Unity was investigat... more In this paper, the concept of using agent behaviors in the 3D environment of Unity was investigated after an introduction to the first cases of Game AI Research in history. The Machine Learning Agents Toolkit was analyzed, and its structure with different use cases was explained. The limitations of using Unity Engine as a platform of Game AI Research and their effects on AI Agents Decision Making Ability were discussed. Finally, possible solutions to overcome these limitations were evaluated. It was concluded that Unity was a solid Platform for Game AI Research and its drawbacks could be nullified by specified methods.
Drafts by Ertugrul Akay
This paper summarises the findings and the theories behind the Game Artificial Intelligence (AI) ... more This paper summarises the findings and the theories behind the Game Artificial Intelligence (AI) project: Connect Four is used as a game domain to train and generate game-playing AI agents using reinforcement learning tech-niques like N Step Look Ahead and random heuristics in the platform of Kaggle collab notebook. The main idea is to create a well-trained agent which is adap-tive enough to play against a variety of agents that have different decision-making capabilities. It is hypothesized that a Reinforcement Learning (RL) agent can be trained to become smarter than its trainee. An experiment is con-ducted to train RL agents and these trained agents are then compared against their trainees in thousand game rounds. The experiment results showed that having the same training environment for differently trained RL agents does limit their potential when their trainee complexity is increased.
Open world games create a refreshing gaming experience by providing a dynamic environment with ma... more Open world games create a refreshing gaming experience by providing a dynamic environment with many interactable game systems and characters. These systems require special game artificial intelligence (AI) agents to create an illusion of an authentic, living world. The hypothesis is, By modeling game systems that are suitable for Multi-Agent Reinforcement Learning and Evolutionary Game Theory principles, it is possible to create game AI agents that can act rationally and adaptive to the dynamic environment. Doing so will hypothetically make open-worlds more compelling and immersive by making players think that they are in a world with intelligent living beings that have already existed long before. The focus will be on creating a single system with adaptive human-like game AI agents first, then gradually introducing multiple game systems and real players. Finally, the results of this research will be used in commercial open-world games and its effect will be observed.
Modern Board games are great mediums for game Artificial Intelligence (AI) research since they in... more Modern Board games are great mediums for game Artificial Intelligence (AI) research since they include complex problems to solve within a set of rules and have various interactable components. Tabletop Games Framework is used to digitalize and implement the board game called Battlelore: Second Edition for this purpose. It is the first wargame to have in the framework and an interesting challenge for different game AI agents. This paper summarizes the process of designing and implementing the board game and numerous gameplaying agents like Rule-Based, One Step Look Ahead, Monte Carlo Tree Search (MCTS) and Rolling Horizon Evolutionary (RHEA) algorithms. After the implementation of the core game with some minor constraints in main features and components, the performance of implemented agents is observed through a Round Robin Tournament. The findings from this experiment are in correlation with the hypothesis for some agents, except for one outperforming (MCTS) and one underperforming agent (Rule-Based). Further work is possible to have betterperforming agents, or some of the game parameters can be tweaked to use the game in different research areas.
In this paper, the concept of using agent behaviors in the 3D environment of Unity was investigat... more In this paper, the concept of using agent behaviors in the 3D environment of Unity was investigated after an introduction to the first cases of Game AI Research in history. The Machine Learning Agents Toolkit was analyzed, and its structure with different use cases was explained. The limitations of using Unity Engine as a platform of Game AI Research and their effects on AI Agents Decision Making Ability were discussed. Finally, possible solutions to overcome these limitations were evaluated. It was concluded that Unity was a solid Platform for Game AI Research and its drawbacks could be nullified by specified methods.
This paper summarises the findings and the theories behind the Game Artificial Intelligence (AI) ... more This paper summarises the findings and the theories behind the Game Artificial Intelligence (AI) project: Connect Four is used as a game domain to train and generate game-playing AI agents using reinforcement learning tech-niques like N Step Look Ahead and random heuristics in the platform of Kaggle collab notebook. The main idea is to create a well-trained agent which is adap-tive enough to play against a variety of agents that have different decision-making capabilities. It is hypothesized that a Reinforcement Learning (RL) agent can be trained to become smarter than its trainee. An experiment is con-ducted to train RL agents and these trained agents are then compared against their trainees in thousand game rounds. The experiment results showed that having the same training environment for differently trained RL agents does limit their potential when their trainee complexity is increased.
Open world games create a refreshing gaming experience by providing a dynamic environment with ma... more Open world games create a refreshing gaming experience by providing a dynamic environment with many interactable game systems and characters. These systems require special game artificial intelligence (AI) agents to create an illusion of an authentic, living world. The hypothesis is, By modeling game systems that are suitable for Multi-Agent Reinforcement Learning and Evolutionary Game Theory principles, it is possible to create game AI agents that can act rationally and adaptive to the dynamic environment. Doing so will hypothetically make open-worlds more compelling and immersive by making players think that they are in a world with intelligent living beings that have already existed long before. The focus will be on creating a single system with adaptive human-like game AI agents first, then gradually introducing multiple game systems and real players. Finally, the results of this research will be used in commercial open-world games and its effect will be observed.