Building Controllers for Tetris

International audienceThis article has two purposes: a review on the problem of building a controller for the well-known video game Tetris, and a contribution on how to achieve the best performance. Key components of typical solutions include feature design and feature-weight optimization. We provide a list of all the features we could find in the literature and in implementations, and mention the methods that have been used for weight optimization. We also highlight the fact that performance measures for Tetris must be compared with great care, as (1) they have a rather large variance, and (2) subtle implementation choices can have a significant effect on the resulting scores. An immediate interest of this review is illustrated. Straightforwardly gathering ideas from different works may lead to new ideas. We show how we built a controller that outperforms the previously known best controllers. Finally, we briefly discuss how this implementation allowed us to win the Tetris-domain prize of the 2008 Reinforcement Learning Competition