Rubik's Cube Robot: From Budapest to ABB YuMi

A Rubik's Cube robot has to read the scrambled colours and then turn six faces fast without jamming the cube or dropping it. This post follows the puzzle from Ernő Rubik's workroom in Budapest to the speedcubing records set by children this autumn, then to machines that solve it in about a third of a second and robot hands that learned to turn it by feel. Along the way we look at what dexterous industrial robots such as ABB's dual arm YuMi have in common with these solvers, and where they differ.

A scrambled 3x3x3 Rubik's Cube. Photo: Rama / Wikimedia Commons (CC BY-SA 2.0 FR)

How a Budapest puzzle turned into a race against the clock

Ernő Rubik, a Hungarian sculptor and professor of architecture, built the first cube in Budapest in 1974. It was first sold there in 1977 as the Magic Cube, and Ideal Toy Corp took it worldwide in 1980 under the name Rubik's Cube. The standard 3x3x3 cube can be mixed into more than 43 quintillion arrangements. Rubik himself is said to have needed about a month to solve his own invention the first time. For years every solve was done by hand, by people working out sequences of moves on paper and passing them around in books and magazines.

Ernő Rubik, inventor of the cube, at the Liberty Science Center Genius Gala in 2014. Photo: Babak Mansouri / Wikimedia Commons (public domain)

The first world championship was held in Budapest on 5 June 1982, and Minh Thai, a Vietnamese student from Los Angeles, won with a time of 22.95 seconds. Competitive solving faded after the craze of the early 1980s and came back with the internet, when solvers shared methods online. Today the World Cube Association runs official events with stackmat timers, scrambles generated by software and strict rules on inspection time, so records from different countries can be compared fairly.

Speedcubers with stackmat timers at the Aachen Open competition in 2010. Photo: heipei / Wikimedia Commons (CC BY-SA 2.0)

Machines joined in once cheap cameras and motors arrived. Hobbyists built solvers from Lego Mindstorms kits, and school clubs printed their own frames. In Koblenz, Germany, students at the Dr. Zimmermannsche Wirtschaftsschule built a cube robot with their teacher, with parts made on the school's 3D printer; in late 2019 it was still being tested and solved more slowly than a good human. The approach in these builds is the same: a camera reads each face, software computes a short solution, and a motor on each face turns it.

A cube solving robot built by students in Koblenz with 3D printed parts, photographed in 2019. Photo: Lothar Spurzem / Wikimedia Commons (CC BY-SA 2.0 DE)

Rubik's Cube records, robot hands and where ABB fits today

Humans keep getting faster, and the newest champions are very young. In September 2026 Lian Yunzhi, a six year old from China, set the women's world record for the 3x3x3 average with 4.52 seconds in Wuhan, then cut it to 4.27 seconds in Guangzhou a few days later, according to the World Cube Association listing reported by Global News and ITV. In early October, at the Guangzhou Grand Open 2026, Xuanyi Geng lowered the single solve world record to 2.51 seconds.

A blindfolded solve at the Estonian Open speedcubing competition in 2016. Photo: Ivo Kruusamägi / Wikimedia Commons (CC BY-SA 4.0)

Robots are much faster. In 2018 a machine built by MIT students Ben Katz and Jared Di Carlo solved a cube in 0.38 seconds. In May 2024 Mitsubishi Electric engineers in Hyogo, Japan, set a Guinness World Record of 0.305 seconds with TOKUFASTbot. Its motors turn a face 90 degrees in 0.009 seconds, and an AI colour recognition step reads the cube first. The team said red and orange squares and shadows from the robot's own grippers were the hard part.

A modern GAN speedcube, the kind of puzzle used in record attempts today. Photo: 多多123 / Wikimedia Commons (CC BY 4.0)

Record machines grip the cube with a motor on every face. In October 2019 OpenAI showed a human-like robot hand, a Shadow Dexterous Hand, solving the cube with one hand after training neural networks entirely in simulation. ABB's YuMi IRB 14000 is somewhere in between: it has two seven axis arms and was built to handle small parts next to people. ABB has not built a YuMi cube solver that we know of, but the same skills of reading a part with a camera and turning it precisely are what YuMi does in electronics and watch assembly, as our ABB robot use cases page shows.

A Shadow Dexterous Hand holding a light bulb, the same make of hand OpenAI trained to solve the cube. Photo: Richard Greenhill and Hugo Elias / Wikimedia Commons (CC BY-SA 3.0)

What cube solving robots could teach factories and homes next

The cube is useful as a test because success is easy to check and the moves are hard to do by feel. OpenAI chose it for that reason, and it reported that its hand could cope with surprises it never saw in training, such as being nudged by a stuffed toy giraffe. That kind of robustness is what factory engineers want when parts arrive slightly out of place. A cube robot that can recover from a slipped face is practising the same thing a robot does when it reseats a connector that did not click.

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

For factories, the realistic next step is two armed robots with cameras doing short manipulation tasks that used to need human fingers. A YuMi or a single arm YuMi IRB 14050 could, in principle, be programmed to rotate a cube face by face with one arm holding and the other turning, and that would make a good teaching exercise. Such a cell would normally be built and tested first in RobotStudio, where grip positions and turn speeds can be tried without wearing out a real cube.

An engineer programs a single arm ABB YuMi at a work table. Photo: ABB

At home, cube robots will probably stay as toys and learning kits, and as a challenge for school robotics clubs. The open problems are the same ones home robots face everywhere: cost, safety around children, and hands that are strong enough to turn a stiff cube but gentle enough not to break it. General purpose robot hands still cost far more than a dedicated solver, so a family that wants a fast cube robot will buy a box with six motors. Robots that turn the cube with human-like hands are still slower than both those boxes and the children now setting world records, and closing that gap is mostly a question of better touch sensing and cheaper hands.

Students working with ABB YuMi robots in an Italian school lab. Photo: ABB

Innovation

AI solutions for effortless ABB robot control.

Automation

Robotics

ceojohntran@chatwithrobot.net

+84905311611

© 2025. All rights reserved.

qtran1215@gmail.com