The robot that solved a Hungarian cube in two seconds is the product of 60 years of research in the field of artificial intelligence

For decades solving games and puzzles was a neglected subfield of artificial intelligence research. Little by little it became clear that ideas that came up within it helped to make research breakthroughs in the field and vice versa. The algorithm for solving the Hungarian cube is an old one, but only today can it be performed thanks to computing capabilities

About two months ago, the ARM company, whose chips are the basis of many cell phones and tablets, asked to demonstrate the speed of Samsung's chips based on its technology and which drive the eight cores of the Galaxy S4 phones (which has meanwhile been replaced by the S5) by solving the three-dimensional puzzle known to all of us as " Hungarian Cube" (Rubik's Cube).

ARM principal engineer David Gilday and his partner Mike Dobson, a security systems engineer at Securi-Plex. Responsible for the Cubestormer 3 robot, which was designed to solve the cube in an impressive time of 3.253 seconds. The two will work on the project in their free time. They were also responsible for the previous record of 5.27 seconds achieved by a previous robot – Cubestormer 2 in 2011.
Cubestormer 3 is powered by the octa-core version of the Galaxy S4 device, which contains a pair of Cortex-A7 CPUs and four Cortex-A15 processors. The device looked at the cube, calculated the steps required to solve it and sent the instructions to four robotic arms. The device used ARM9 processors in eight Lego Mindstorm EV3 bricks, which helped control and sequence the actions.

"It's trickier than it looks because Cubestormer 3 uses a quick die that allows it to be flipped before both sides are completely lined up." Gilday said after the feat. This means that the robot effectively reflects the same kind of judgment and skill that human cube solvers would exercise.”

The world record for solving a human Hungarian cube stands at 5.5 seconds and is held by Mats Valk from the Netherlands. "Although the time is relatively close to the record of Cubestormer 2 from 2011, it is significantly far from the new record, and if there is no significant change in human evolution, how likely is it to be broken in the near future." said.
Of course, other challenges remain for humans, for example the record for solving a 7 x 7 cube is 2 minutes and 39 seconds.
"We knew that the CUBESTORMER 3 robot had the potential to break the Guinness World Record and the achievement of its previous generation, although there is some risk that the physical actions of the robot are faster than the human eye is able to perceive," notes Gildai. "In the end, the many hours spent developing the robot and the constant improvement in its motor and technological capabilities paid off. The significant challenge facing us right now is to try and improve the speed of the robot even more."

The Hungarian cube is an experimental field for the heuristic research which is a subfield of artificial intelligence. The Hungarian cube has 10 to the 19 possible states, making them impossible to store in memory.

There are two common ways of setting steps in a Hungarian cube. The first definition refers to the step as a quarter turn (90 degrees) of each side. The second definition looks at a step as a rotation of the face of the cube, whether it is 90, 180 or 270 degrees.

The Hungarian cube was invented in the late seventies and became a popular toy in the eighties. By 1982 it had already become a focus of interest for researchers of Malakhotzi intelligence. In 1997, Richard Korff showed that the minimum number of steps required to solve a randomly arranged Hungarian cube is at least 20.
In 2010, several researchers showed that any arrangement of the Hungarian cube can be solved in no more than 20 steps. This calculation work required 35 years of CPU processing. This ended their 30 years of research in the field of artificial intelligence dedicated to the search for the minimum and maximum steps required to solve a Hungarian cube.

in his article "Play with AI" Haim Hirsch of Rutgers University writes: The use of puzzles and games in artificial intelligence research dates back to the early days of the field. In the early fifties, Claude Shannon and Alan Turing wrote an article proposing to create computer programs that could successfully handle games that require two players. Most of the researchers in the field, most of them from IBM's research laboratories, preferred chess and others suggested to be satisfied with checkers (Arthur Samuel at IBM). But some also offered to solve puzzles. In the early XNUMXs, Navel, Shaw and Simon used some puzzles in their GPS software (no relation to the navigation system).

A puzzle is a different problem to solve using automated approaches. Although in the sixties, puzzles and games were seen as an important experimental field for artificial intelligence research, the games that require cognitive skill were seen as too difficult to invest in research, but they were used as an experimental field for explaining basic questions in artificial intelligence. For years, the work on puzzles and games was isolated from the rest of the research in the field of artificial intelligence, but this has recently begun to change. Many ideas developed in other areas of artificial intelligence have proven to be of practical importance to those working on game and puzzle solutions.

Similarly, the problems that have been unfairly labeled as the 'toy domain' have proven to be fruitful areas for AI law and have contributed ideas to other areas of AI. Reasonable and correct use of approximation can be important, as in Korff's evaluation functions for the Hungarian cube based on the study of simplified versions of the problem. Adding Korf's evaluation functions to solving the Hungarian cube requires enormous computing resources to store tables that can only be found today in computerized storage systems.