Arabian nights: learning agent in a game environment

• Main Authors: Abada, Ronald Van Andrew C., Ang, Albert Eugene T., Kimseng, Karl James N.
• Format: text
• Language: English
• Published: Animo Repository 2004
• Subjects: Computer games > Design; Computer algorithms; Machine learning; Computer Sciences; Theory and Algorithms
• Online Access: https://animorepository.dlsu.edu.ph/etd_bachelors/6603
• Institution: De La Salle University

Arabian Nights is a system simulated as a computer management game populated by an autonomous learning agent and several reactive agents. The agents in the game are able to interact with their environment and with one other. By observing the actions of the player, the learning agent learns how the player plays the game. The learning agent makes use of case-based reasoning to retrieve the actions of the user in a previous game situation most similar to the current game situation. Revision is then used to adapt the retrieved set of actions to the current games situation when some actions cannot be done. This enables it to mimic the actions of the player and to make its own decision in the game based on the previous decision of the player. The probability of a set of actions being retrieved depends on the evaluation done to it by the player and the learning agent. The player can evaluate the set of actions the learning agent executes. The learning agent also has the ability to evaluate the actions it executes. It does self-evaluation by comparing the resources of a previous game situation to the current game situation. An increase in resources gives a positive self-evaluation to the set of actions it retrieved while a decrease in resources does the opposite. The higher the evaluation of a set of actions, the higher the probability it has in getting retrieved. The reactive agents in the game interact with buildings to produce resources. The research emphasizes on the learning of an agent.