MarioMix: Creating Aligned Playstyles for Bots with Interactive Reinforcement Learning
This addresses the challenge for game designers without ML expertise to customize bot behaviors, though it is incremental as it applies existing interactive RL methods to a specific domain.
The authors tackled the problem of enabling non-expert game developers to create bot behaviors with aligned playstyles by proposing MarioMix, a tool based on interactive reinforcement learning, which allowed designers to select preferred playstyles from precomputed bots for Super Mario Bros.
In this paper, we propose a generic framework that enables game developers without knowledge of machine learning to create bot behaviors with playstyles that align with their preferences. Our framework is based on interactive reinforcement learning (RL), and we used it to create a behavior authoring tool called MarioMix. This tool enables non-experts to create bots with varied playstyles for the game titled Super Mario Bros. The main interaction procedure of MarioMix consists of presenting short clips of gameplay displaying precomputed bots with different playstyles to end-users. Then, end-users can select the bot with the playstyle that behaves as intended. We evaluated MarioMix by incorporating input from game designers working in the industry.