Developer Kevin Hughes says he's managed to train an artifical intelligence agent to play Mario Kart 64, and last month he published a blog post walking through how he did it -- plus some lessons learned along the way.
It's a neat project (you can see the successful outcome above) and game devs should note that Hughes has shared his "TensorKart" code on Github, along with the driver he built to feed input data into an N64 game emulator.
The full (and brief) blog post is well worth reading, but here's a quick summary: Hughes says he relied on open-source machine intelligence software TensorFlow for this project, eventually turning to Nvidia Autopilot -- the company's implementation of TensorFlow for training actual self-driving cars -- to train his automated neural network to navigate a track in Super Mario 64 running on an emulator.
"In a couple of days over winter break, I was able to to train an AI to drive a virtual vehicle using the same technique Google uses for their self-driving cars," wrote Hughes, who currently works as a developer at Shopify. "With approximately 20 minutes of training data, my AI was able to drive the majority of the simplest course, Luigi Raceway, and generalized to at least one section of an untrained racetrack. With more data, I bet we could build a complete AI for Mario Kart 64."