LEGO Sorting Machine Uses Neural Networks to Sort Any Piece

Anybody who has a penchant for LEGO probably has a bucket or two of messily mixed brick somewhere in their abode—like sands through the hourglass, so are the pieces of sets once destroyed by los gatos. But it’s the 21st century, which means sorting those myriad pieces by hand could soon be a thing of the past. If AI-powered LEGO sorting machines like the one featured here manage to catch on, that is. (It could happen, just look at what the Turing machine eventually turned into.)

Daniel West’s universal LEGO sorting machine. 

Gizmodo picked up on this “World’s First Universal LEGO Sorting Machine,” which was built by software engineer, AI enthusiast, and game developer Daniel West. West, who has a YouTube channel with a handful of fun, LEGO-centered videos like this one demonstrating a LEGO “Bolt Action Rifle with Grenade Launcher,” says he spent two years building this universal LEGO sorting machine. And while spending two years on anything related to LEGO seems like a recipe for madness—especially if it gets knocked over—something this complex obviously requires that kind of epic effort.

Speaking of epic effort, West’s universal sorter is itself built out of 10,000 LEGO bricks, which is awesome not only because it demonstrates his ingenuity, but also the limitless possibilities of building with LEGO. Six LEGO motors and nine servo motors power the machine, working in unison to move various conveyor belts, sorting gates, and even a “vibration feeder.”

This LEGO sorter using artificial intelligence to sort thousands of different LEGO types.

Neutral networks sort the individual pieces discharged by the “vibration feeder.”

The most astounding part of the universal LEGO sorter is its use of artificial intelligence, specifically machine learning, and even more specifically, convolutional neural networks. West describes in more detail how the neural networks work in the video below, but it’s more or less all about big data and pattern recognition.

The neural networks deployed by the universal sorter have been trained on massive data sets made up of all the various LEGO pieces, which allows them to identify any piece that makes its way through the machine. As West points out, however, there weren’t nearly enough real-world images of the multitude of LEGO piece types, so he needed to train the neural networks using digital images instead.

West’s video explaining how the neural networks deployed by his LEGO sorting machine work. 

Thanks to the well-trained neural networks, as well as the efficiency of West’s machine, the universal LEGO sorter is able to sort at a rate of one brick per every two seconds. The machine is able to identify 3,000 different brick types, as well as bricks it hasn’t seen before, so even the bizarre on-off pieces can be sorted. Although the sorter doesn’t seem to be able to identify colors yet, so there’s still room for improvement.

What do you think of this AI-powered universal LEGO-sorting contraption? Do you think a mass-market version of this dream sorting machine will ever be sold on store shelves? Sort out your thoughts in the comments!


Images: Daniel West