SmartAgent - Creating Reinforcement Learning Tetris AI

From Meritology

Author's Abstract

For an NP-complete problem with a large amount of possible states, such as playing the popular videogame of Tetris, learning an e�ective arti�cial intelligence (AI) strategy can be hard using standard machine learning techniques because of the large number of examples required. Reinforcement learning, which learns by interacting with the environment through actions and receiving reward based on those actions, is well suited to the task. By learning which actions receive the highest rewards, the AI agent becomes a formidable player. This project discusses the application of reinforcement learning to a Tetris playing AI agent which was entered into the Reinforcement Learning Competition 2008 and presents the results and conclusions formed from its development and performance.

---

---

Resource: [1] Title: SmartAgent - Creating Reinforcement Learning Tetris AI Author: Samuel J. Sarjant