Researchers Develop System to Train AIs Using Mega Man II
Video games have long been a training ground for AI, so its not much of a surprise that researchers are using the classic Nintendo game Mega Man II to develop AI agents.
Researchers are teaching their AI to play Mega Man II
Video games have been used as a way to train AI algorithms as long as there have been AI algorithms since they provide a dynamic environment that researchers can use to test the efficacy of the systems that they are building. According to a new paper [PDF] posted to the ArXive preprint server, the 1988 classic Capcom game Mega Man II is the latest game to become an AI training ground.
An international team of researchers have developed EvoMan, a modified AI testing framework based on the eight boss fights of Mega Man II with the goal of having an AI agent trained to defeat each boss and have their performance measured against a baseline value.
What makes EvoMan different from other similar challenges is that rather than have a single enemy to test their AI against, EvoMan features eight bosses with a twist. Each boss has a unique weapon that EvoMan can use once the boss is defeated that makes certain remaining boss fights easier. The goal is for an AI agent to perform equally well against each boss.
The researchers have proposed a competition to see whose AI agent performs the best in their framework: "The winner of this competition will be the one agent that performs equally well on each one of the eight bosses, hopefully defeating them all."