Jigsaw Puzzle Robot: From Spilsbury Maps to ABB Arms

A jigsaw puzzle robot sounds like a gimmick until you watch one try to seat a single piece. The job mixes vision, touch and patience, which is why researchers keep coming back to it as a test for robot hands. This post follows the puzzle from John Spilsbury's cut maps in London around 1760 to hydraulic die cutting, then to the small arms that now solve puzzles on tabletops. It covers the news from the past few weeks, shows where ABB arms such as YuMi and GoFa could fit, and ends with the open problems for homes and factories.

A hand fitting a piece into a half finished jigsaw puzzle on a wooden table. Photo: kallerna / Wikimedia Commons (CC BY-SA 4.0)

How a London mapmaker's saw made the first jigsaw

John Spilsbury, a London engraver and mapmaker, is generally credited with the first jigsaw puzzle, made around 1760. He mounted a map on a sheet of hardwood and cut along the borders of the countries with a marquetry saw. He sold the pieces as "dissected maps" to teach geography, since a child learned where each country sat by fitting it back into place. The British Library holds one of his maps of Europe, dated 1766. The toy began as a school lesson, and every piece was cut by a craftsman with a fine saw.

A map of Europe cut into pieces by John Spilsbury in 1766. Photo: British Library via Wikimedia Commons (CC0)

Dissected maps stayed popular for decades. J. Wallis published them for English counties and continents in the early 1800s, and McLoughlin Brothers issued a dissected map of the United States around 1900. The name followed the tool: the fretsaw became the standard cutting tool around 1880, and the phrase "jigsaw puzzle" is in print by 1906. Hand cutting was slow, and the cardboard puzzles that appeared in the late 1800s replaced wooden ones only gradually. Whoever wanted a puzzle still depended on someone with a saw and a steady hand.

An English county map of Somerset cut as a dissected puzzle by J. Wallis in 1812. Photo: Library of Congress via Wikimedia Commons (public domain)

The machine step came in the 1930s, when hydraulic presses pushed printed cardboard through steel blades in a puzzle die. One stroke now cut a whole picture, and puzzles became a cheap pastime that sold well through the Great Depression. Robots arrived at the table much later. A 2023 study in the journal Sensors used a Dobot Magician desktop arm with a suction cup and two cameras to rebuild puzzles of 35 and 70 pieces. The authors reported an average success rate of 87.1 percent, and accuracy fell as the piece count rose.

A Dobot Magician desktop robot arm, the model used in a 2023 jigsaw solving study. Photo: Giacomo Alessandroni / Wikimedia Commons (CC BY-SA 4.0)

Speed solvers, record attempts and a robot that wiggles

Puzzles are having a visible moment. The New York Times ran a guide to puzzles worth the table space on October 7, and on October 12 Fortune covered a maker of $950 wooden jigsaws. The New York Jigsaw Puzzle Association scheduled a Guinness record attempt for September 26 in Binghamton, New York. Competition is organised as well: the World Jigsaw Puzzle Championship seats solvers at rows of identical tables for timed rounds. Human solvers set the speed that a robot has to beat.

Rows of identical tables seen from above at the 2023 World Jigsaw Puzzle Championship. Photo: World Jigsaw Puzzle Federation / Wikimedia Commons (CC BY-SA 4.0)

In 2024 Mark Rober's team built a robot to race Tammy McLeod, one of the faster human solvers. Early runs had many misfits, so the team fixed errors in the computer vision code. They also copied a human habit. People wiggle a piece until it snaps into place, and a robot that only presses straight down misses the fit, so the team programmed a wiggle. A 2020 benchmark called DeepClaw hit another snag: its puzzle piece was only 5 mm thick, too thin for a gripper, so the cells used suction.

A competitor leans over a large puzzle at the 2022 World Jigsaw Puzzle Championship. Photo: SemiEvolved / Wikimedia Commons (CC BY-SA 4.0)

I found no ABB jigsaw puzzle project, so this part is a possibility. The dual arm YuMi IRB 14000 is made for small part assembly, and the GoFa CRB 15000 can be taught by guiding it with a hand. The IRB 1700 family, which ABB launched at CIIF in Shanghai in mid October, lists a path repeatability of 0.03 mm. ABB also released an AI vision tool on October 16 that it says cuts commissioning time by up to 90 percent. A puzzle cell would be built and tested first in RobotStudio before any piece is moved.

An ABB YuMi assembling small parts beside a lab worker. Photo: ABB

What a puzzle robot teaches homes and factories

For a home, the likely robot is a tabletop arm that sorts a fresh box into edge pieces and colour groups, leaves the satisfying middle to you, and fills in a stubborn corner when asked. No source I found shows a consumer puzzle robot on sale. Cost is the first barrier, because a camera, an arm and software must sell for less than the pleasure of doing it yourself. Safety is the second, since an arm beside a child's hand needs force limits, which is the reason cobots such as GoFa stop on contact.

A thousand piece puzzle in progress on a table, with spare pieces in a tray. Photo: Balise42 / Wikimedia Commons (CC BY-SA 4.0)

The technical problems are specific. Pieces are thin, and a suction cup or finger must lift one without disturbing its neighbours. Fit tolerance is tight, because a die cut piece seats only when position and angle are right within a fraction of a millimetre. Many puzzles also have large areas of similar colour, so matching shapes alone is not enough and the system has to read texture as well. A 2025 review noted that few papers address robotic puzzle assembly at all, so the field is still open for students.

The new generation ABB IRB 1200, a compact six axis arm for fast small part work. Photo: ABB

The skills transfer to factories. Finding a part with a camera, nudging it until it seats, and checking the result are the same steps used in snap fit assembly and peg in hole work. Readers who want to see where those steps are already used can start with the ABB robot use cases, and the Rubik's cube robot post covers a related puzzle. Skills are the open problem there: someone has to set up the vision, tune the force limits and keep the cell running.

An ABB GoFa cobot loading parts into a machine at a small factory. Photo: ABB

Innovation

AI solutions for effortless ABB robot control.

Automation

Robotics

ceojohntran@chatwithrobot.net

+84905311611

© 2025. All rights reserved.

qtran1215@gmail.com