Machines may well be on the road to world domination, but one programmer decided that they should make a pit stop on the way: Mastering Mario Kart. YouTuber and programmer SethBling recently unveiled his newest creation, MariFlow, which uses machine learning to create a bot that can navigate Mario Kart courses like a human.
In a video titled, “How I trained a recurrent neural network called MariFlow to play Super Mario Kart like a human,” SethBling explains that MariFlow is a neural network (a type of artificial intelligence) that can learn by example. He trained the network to mimic his own gaming style by carefully examining 15 hours of gameplay to understand how he used the controller to drive. Ultimately, MariFlow’s goal is for the AI’s driving to be indistinguishable from a human.
So far, the network is winning! MariFlow drives on its own and wins gold medals in 50cc Mushroom and Flower Cups, but its creator admits that it has only managed to place second in Star Cup.
This is not the programmer’s first cool self-learning gaming project, either. In the same video, SethBling discusses another neural network, MarI/O, which learned how to master playing Mario World on its own. Unlike MariFlow, MarI/O’s game play evolved on its own. It produced small networks with evolved connections, which resulted in gameplay that could leave any human in the dust.
Want to help MariFlow learn how to play even better? SethBling released the MariFlow code and all of his recorded inputs so other players can help train it to whip around courses like a whiz.
H/T Kotaku
