Mahjong Robot: From Hand-Built Walls to ABB Arms

A mahjong robot has to pick up small tiles, read them, decide, and discard without knocking over the wall. Machines already do part of that. Automatic tables have shuffled tiles and built the walls in Japanese parlors since the 1970s, and a Chinese company showed a robot playing a hand in 2025. This post follows the game from hand-built walls to automatic tables, looks at the 2026 mahjong boom and the first robot players, and asks where small ABB robots such as YuMi and GoFa could fit.

A Japanese mahjong set with its racks and counting sticks, the 136 tiles every table must shuffle. Photo: Aleksey Zhukov / Wikimedia Commons (CC BY-SA 2.0)

How mahjong was shuffled and stacked by hand

Mahjong took shape in China in the late Qing era, and the earliest surviving tile sets date to about 1870. Ningbo and Fujian are both named as its birthplace. British consul F.E.B. Harvey described learning the game in Ningbo in the 1860s. For players, the work was physical. Every round, the four of them pushed all 136 tiles face down, washed them with both hands, and built four walls by hand. Each hand began with several minutes of clacking and counting before anyone could play a tile.

Tiles stacked into walls on a felt table, the job players once did entirely by hand. Photo: yui* / Wikimedia Commons (CC BY-SA 2.0)

The game reached the United States in the 1920s. Joseph Park Babcock published Rules of Mah-Jongg, the "red book", in 1920, a simplified version for American players. Abercrombie and Fitch began selling sets that year and sold about 12,000 in all, with co-owner Ezra Fitch sending buyers to Chinese villages to find them. By 1924 the craze had reached a bathing beach, as a National Photo Company picture shows. In 1937 the National Mah Jongg League was founded to standardize American rules, starting with 32 members.

Four swimmers play mah-jong on a floating table at a bathing beach in 1924. Photo: National Photo Company via Library of Congress and Wikimedia Commons (public domain)

Japan is where the shuffling was first mechanised. Japanese patent filings for devices that push and arrange tiles date from the late 1970s, and ABEMA reports that the first fully automatic table appeared in 1976, during a mahjong boom, and went into many parlors. Players soon complained that the tiles were mixed poorly. Taiyo Giken, a late entrant, fixed that with better mixing and tables that line up each player's dealt tiles, and it is now the largest maker. Its AMOS REXX III was announced in October 2020. By Wikipedia's count Japan had about 8,900 parlors and 7.6 million players in 2008.

A mahjong parlor sign with the characters for mahjong and a parking arrow in Japan. Photo: 円周率３パーセント / Wikimedia Commons (CC BY-SA 4.0)

Tile factories, home tables and the first robot players

The latest boom is American. Sixth Tone's report on the Chinese factory town of Taixing says Yelp recorded a 4,467% rise in U.S. searches for mahjong clubs between September 2024 and August 2025, and the National Mah Jongg League now has more than 350,000 members. Nearly 30 new factories opened in Taixing in a year. One owner spent 1 million yuan on new equipment and now makes up to 200 American sets a day, up from about 20 sets a year. American tiles are smaller and carry more detail, so one sorter's daily output of sets dropped from about 150 to about 60.

Players mix tiles by hand at a mahjong day with students at the University of North Georgia. Photo: Huntsmanleader / Wikimedia Commons (CC0)

Automatic tables are also moving into homes. In May 2018 Sunco released a folding automatic table in Japan for 69,800 yen with tax, aimed at players worried about noise and floor space. A table shuffles but does not play. In April 2025 a Chinese embodied-AI company, PsiBot, released a video of a robot that calls pong and kong and declares a self-drawn win. Reports say it moves more slowly than a person, and the claims come from the company, so treat it as a demonstration, not a product.

A chess robot arm faces a human player at the Women's World Chess Championship in Sochi, 2015. Photo: Sergei Kazantsev / Wikimedia Commons (CC BY-SA 4.0)

ABB has no mahjong project on record, but its small robots do similar work in factories. The dual-arm YuMi IRB 14000 was built to handle small parts beside people, and ABB has shown a YuMi playing tic-tac-toe with visitors at a robotics event. A tile is a small, stackable part, so a YuMi with a soft pad or suction gripper and a camera could plausibly slide one out of a rack and set it face up. That is a possibility to test, not a finished cell. A team would work out the motions first in RobotStudio.

An ABB YuMi playing tic tac toe with visitors at the FIRST Championship in Detroit, 2019. Photo: Stilfehler / Wikimedia Commons (CC BY-SA 4.0)

What a robot at the mahjong table still has to learn

The table itself is a solved problem, and a camera can already read tile faces. The arm is harder. A hand holds 13 or 14 tiles that must be picked up and turned without disturbing the neighbors, and a discard has to land in the middle without touching the others. A robot also has to cope with scuffed tiles, tiles of different sizes between Japanese, Chinese and American sets, and the smaller designs now coming out of Taixing. Our post on chess playing robots shows what a board game already asks of a gripper.

A family plays mahjong at home on a table with a pink cover. Photo: SwisseRosetree / Wikimedia Commons (CC BY-SA 4.0)

At home, the likeliest first helper is a tabletop machine that deals and keeps score, not a player with arms. A robot that fills an empty fourth seat would need to be safe near children, quiet, and cheap, and today's cobots are priced for factories. ABB's GoFa stops when it feels contact, which is the right idea, but it is far larger than a tile game needs. Anyone who wants to experiment can start by controlling robots with no code on a classroom arm.

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

Trust is still open. Players like to see the tiles washed and want to know nobody can peek, so a robot has to show that in plain view. House rules also differ between Japanese riichi, Chinese and American mahjong, and a robot would have to learn each one before anyone trusted it with a seat. For the factories in Taixing, where margins are falling and competition is rising, robots and AI tools will probably arrive on the sorting and packing lines long before they sit at a family table.

ABB's GoFa cobot is guided by hand and stops when it senses contact. Photo: ABB

Innovation

AI solutions for effortless ABB robot control.

Automation

Robotics

ceojohntran@chatwithrobot.net

+84905311611

© 2025. All rights reserved.

qtran1215@gmail.com