Year: 2015
Type: conference-paper
Abstract: Monte Carlo Tree Search (MCTS) has recently seen considerable success in playing certain types of games, most of which are discrete, fully observable zero-sum games. Consequently there is currently considerable interest within the research community in investigating what other games this algorithm might play well, and how it can be modified to achieve this. In this paper, we investigate the application of MCTS to simulated car racing, in particular the open-source racing game TORCS. The presented approach is based on the development of an efficient forward model and the discretization of the action space. This combination allows the controller to effectively search the tree of potential future states. Results show that it is indeed possible to implement a competent MCTS-based racing controller. The controller generalizes to most road tracks as long as a warm-up period is provided.
Authors: Jacob Fischer, Nikolaj Falsted, Mathias Vielwerth, Julian Togelius, Sebastian Risi
Institutions: New York University, IT University of Copenhagen
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W2577217803
