Abstract
Since the first Mario AI Competition in 2009 where Mario AI Framework made a debut, there is a continuous interest to use the game and the framework for research in AI. Decent amount of these papers is using the A-Star algorithm along with the game forward model and some results are directly depending on the quality of the best known A-Star-based agent by Robin Baumgarten.
In this paper, we are revisiting and improving the implementation of Mario AI Framework's forward model by two orders of magnitude and also presenting a new A-Star-based agent that outperforms the current state-of-the-art agent by a large margin.