Go Playing Robot: From Weiqi Boards to ABB Cobot Arms

A go playing robot needs a brain and a steady hand. See Go's history, AlphaGo and KataGo, and how ABB YuMi and GoFa arms could place the stones.

INDUSTRIAL ROBOTICS

Chat With Robot

10/19/20265 min read

A go playing robot has two jobs that are easy to confuse. One is choosing the move, which software now does better than any human. The other is picking up a smooth stone and setting it on a grid intersection without nudging its neighbours, and that falls to an arm. This post follows Go from ancient China to the AI era, looks at a recent result from Seoul, and asks where ABB's YuMi and GoFa robots could fit if someone wanted a robot opponent at the kitchen table. ABB has no Go product, so that part is a possibility.

Close-up of black and white Go stones on a wooden board
Close-up of black and white Go stones on a wooden board

Black and white Go stones on a wooden board. Photo: Dietmar Rabich / Wikimedia Commons (CC BY-SA 4.0)

How weiqi went from Chinese courts to club tables

Go began in China more than 2,500 years ago and is usually described as the oldest board game still played. The Chinese call it weiqi, roughly "encirclement board game". The Zuo Zhuan, a Chinese chronicle, is generally taken as the earliest written reference, describing an event from 548 BCE. Players put black and white stones on the intersections of a grid, and once a stone is down it stays there unless it is captured. That rule matters later for robots, because a placed stone is never slid around.

White Go stones in a wooden bowl
White Go stones in a wooden bowl

White Go stones heaped in a wooden bowl. Photo: liz west / Wikimedia Commons (CC BY 2.0)

The game travelled. It reached Korea between the 5th and 7th centuries, where it is called baduk, and arrived in Japan in the 7th century. It was popular at the Japanese court by the 8th century and among ordinary people by the 13th. After 1603 the Japanese state recognised official Go schools, among them the Hon'inbo school. Boards settled on 19 by 19 lines, which gives 361 intersections, and a Sui dynasty tomb model already shows a 19 by 19 board.

Four men playing Go on wooden boards at a club table
Four men playing Go on wooden boards at a club table

Players at a club game at the Seattle Go Center in 2007. Photo: Joe Mabel / Wikimedia Commons (CC BY-SA 3.0)

Today the game lives in club rooms and tournament halls as much as on online servers. Players still handle every stone by hand, lifting one from a bowl between two fingers and clicking it onto the wood. A machine that wants to play across a physical board has to copy that small, practised gesture, and it has to do it on a board that may be knocked or shifted by the players. The first serious attempts were laboratory projects. A 2009 paper from Missouri University of Science and Technology described a Lynxmotion hobby arm named Cheonsoo-I that placed stones for a computer player.

A young woman and a crowd of onlookers around a Go board
A young woman and a crowd of onlookers around a Go board

Onlookers gather around a Go game at a tournament. Photo: Leedors527 / Wikimedia Commons (CC BY-SA 3.0)

Why AI mastered Go while arms lag behind

The big shift came from software. In October 2015 AlphaGo beat the European champion Fan Hui 5 to 0, the first time a program beat a professional on a full board without a handicap. In March 2016 it beat Lee Sedol 4 to 1 at the Four Seasons Hotel in Seoul, and Lee won only game four. People often assume an arm played those games. AlphaGo had none. DeepMind team member Aja Huang, an amateur 6-dan player, placed its stones on the board.

Smiling Go player Lee Sedol in a dark jacket
Smiling Go player Lee Sedol in a dark jacket

Lee Sedol, who played AlphaGo in Seoul in March 2016. Photo: LG Electronics / Wikimedia Commons (CC BY 2.0)

The recent news comes from Seoul too. Dexerto and PC Gamer reported that Shin Jin-seo, a nine-dan Korean professional, became the first human to win an official series against the open-source engine KataGo. Shin played with a two-stone handicap, lost game one, won game two, then took the deciding third game in 221 moves by 11.5 points. Even now a top human needs a head start to beat the software. The physical side is easier to see in a chess robot arm, which we covered in our chess robot post.

A red robot arm reaching toward a chess board across from a seated man
A red robot arm reaching toward a chess board across from a seated man

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 does not sell a Go robot, so here is what its existing arms could do. The GoFa CRB 15000 is a collaborative arm that can be taught by guiding it by hand, and it stops when it senses contact, which suits a game played at arm's reach. Its small size also keeps it out of the way of the player's own hands. A camera would read the board, and that is simple in principle because every point is black, white or empty. A cell like this would normally be built and tested virtually first in RobotStudio.

A white and grey ABB GoFa cobot arm beside a work table in an office
A white and grey ABB GoFa cobot arm beside a work table in an office

An ABB GoFa CRB 15000 cobot at a work table. Photo: ABB

What a Go robot would take at home and in the factory

The YuMi dual-arm robot looks like a better match for fine placement. YuMi IRB 14000 was designed for small-parts assembly beside people, and a stone is a small part. One arm could hold a bowl while the other places stones or clears captured ones. Placing without disturbing neighbours is the hard part, since a stone nudged by a millimetre changes nothing in the rules but looks sloppy to any player. Nobody has shown ABB hardware doing this, so treat it as an idea.

A woman guiding an ABB GoFa arm by hand at a table
A woman guiding an ABB GoFa arm by hand at a table

Lead-through teaching: moving an ABB GoFa arm by hand. Photo: ABB

For homes, the open problems are cost and safety. A cobot arm costs far more than a Go board and a human opponent, and KataGo-class software already runs on an ordinary computer. A physical arm would be wanted for the click of the stone and for an opponent you can look at across the board. A child will reach over the board, so safety matters. Force-limited arms like GoFa are built for contact, but a home installation would still need a proper risk assessment, and nobody should assume a toy price.

Three ABB GoFa cobot arms of different sizes on a grey background
Three ABB GoFa cobot arms of different sizes on a grey background

ABB's GoFa cobot family, arms sized for work next to people. Photo: ABB

For factories, the skills a Go arm needs are the ones assembly cells need every day: seeing a small part, picking it from a loose pile and setting it down without disturbing what is already there. Programming those motions is getting easier, and teaching robots by demonstration shows where that is heading. A Go arm would be a modest demonstration of that progress, and it would give people a friendlier first meeting with a cobot than a factory floor does. Cost, safety checks and the camera work are the same questions an engineer asks on any new cell.

Black and white stones spread over a light wooden Go board
Black and white stones spread over a light wooden Go board

Stones scattered across a 13 by 13 Go board mid game. Photo: Dietmar Rabich / Wikimedia Commons (CC BY-SA 4.0)

Innovation

AI solutions for effortless ABB robot control.

Automation

Robotics

ceojohntran@chatwithrobot.net

+84905311611

© 2025. All rights reserved.

qtran1215@gmail.com

Instagram