Puzzles are a lot of fun, and completing them requires a surprising amount of problem-solving skills that humans are uniquely suited to. With a typical puzzle, the printed image acts as a key reference to determine the location of the pieces. But there are special puzzles that are all one solid color, forcing solvers to find the placement of pieces based solely on the unique shape of the edges. Such puzzles require months of trial and error to solve, and that’s for a human. Building a robot to solve those solid color puzzles it’s quite a challenge, but Shane Wighton’s engineering skills were up to the task.
As Wighton says, this is the future and robots should be able to have fun for us. But this is still quite an engineering challenge. To explain how Wighton made this work, we’ll start with the simplest part of the equation: the motion system. The robot resembles a large CNC router and that’s pretty close to what it is in practice. The sturdy wooden table supports a CoreXY kinematic motion system, which enables very fast movement. You can pick up individual jig saw parts using a vacuum suction end effector like you would find on a pick and place machine. A vacuum suction table, like the one used to clamp plywood on CNC routers, holds parts in place after the robot drops them.
All of that is very impressive and it was easy compared to the rest of the robot. In order for this robot to solve the puzzle, it must compare each edge of each piece with each edge of everything. other part. For a 5000 piece puzzle like this, with 4 sides on each piece, the result is approximately 1.82×1077337 comparisons. If the robot made one comparison per second, it would take 4.2×1077319 times the age of the universe to complete the comparisons. Wighton’s algorithm performs comparisons much faster than once per second, and presumably takes shortcuts like skipping edges and solving pieces, but still estimates that it would take around 3,000 years to solve the puzzle. In his next video, he plans to improve the algorithm and will go into more detail at that time.
Until then, it’s worth understanding how the robot compares parts. Wighton designed a magazine for the robot to grab pieces of. He then places the pieces in a backlit window to take a photo. A normal camera would distort the image and make edge measurements inaccurate, so Wighton used a specialized (and very expensive) telecentric lens. That lens produces an image that appears to be taken from an infinite distance, removing all distortion so that every edge is perfectly perpendicular to the image plane. With that distortion-free image, Wighton could use computer vision software to detect the edges of the pieces and collect precise measurements for his solving algorithm.
To prove it, Wighton made a custom 45-piece puzzle. The robot successfully solved that puzzle, but even this small puzzle took about an hour and a half to complete. The time per additional piece increases exponentially, not linearly. To see how Wighton manages to overcome that daunting hurdle, be sure to check out the next video of him.