Why it matters: For unleashing creativity across all ages, it's hard to beat Lego, the latest proof of which is given by Daniel West, creator of the Universal LEGO Sorting Machine. Combining AI, Lego bricks, motors and a Raspberry Pi, among other things, his creation is able to sort every Lego piece the company has ever made into 18 different buckets depending upon their category, hence the use of 'Universal' in the name.
Are you sick of sorting Legos? West was, apparently. And even if you're not, it's a good idea to take a look at his creation that sorts Lego bricks, while itself being constructed from over 10,000 Lego pieces, 6 lego motors, and 9 servos.
Daniel acknowledges two similar efforts done in the past and refers to his machine as "the next step in this line of evolution." He also says that the Lego sorter can recognize any Lego part ever produced, even those which his machine has never seen before.
Sorting the Legos is a three-phase process that begins by putting a mixed collection of Lego pieces in the input bucket. These are then taken via conveyor belts to a vibration feeder that shakes them off each other to make for a clean and consistent flow into the scanner.
A camera then records a video stream of the Lego piece going through the scanner, after which a Raspberry Pi processes the footage and transmits the images to Daniel's laptop.
Using an AI technique called convolutional neural network, the received images are used to classify the lego piece according to its category and the results are sent back to the machine. From there onwards, a series of gates on the distributor open and close to direct the Lego pieces into one of the eighteen output buckets.
Daniel's machine can identify and sort one Lego brick every two seconds, from the entire catalog of nearly 3,000 different Lego pieces.
Thanks to 3D models, available from sites like Rebrickable, Daniel's AI was able to train at a much faster pace and required him to only use real lego photos for further improving his algorithm. He also credits Blender, TensorFlow and Raspberry Pi for making this project possible.