SpelunkBots API - An AI toolset for Spelunky
Daniel Scales, Tommy Thompson · 2014
This paper describes the Spelunkbots API, a proposed benchmark for computational and articial intelligence algorithms. The benchmark, developed by the authors, is an API designed to allow for AI controllers to be written for the game Spelunky: a challenging 2D platforming game that requires players to commit quick reactive behaviours as well as long-term deliberation in order to maximise reward and minimise completion time. In this paper we highlight the features of the Spelunky game and a rationale for its relevance as an AI benchmark. Followed by a description of how the original Spelunky source code has been modified to provide a testing environment for bots. As well as some examples of simple hand-written bots, we discuss the relevance of this benchmark in the context of current AI methods.