Rock Paper Scissors Robot: Janken to ABB Vision

Rock paper scissors looks like a game of pure chance, so a rock paper scissors robot that never loses sounds like a trick. The best known one simply watches your hand and answers faster than you can finish the throw. This post follows the game from its roots in China and Japan to the University of Tokyo robot hand that wins every round by reading a human hand within a millisecond. After that we look at where high-speed vision shows up today, including ABB robots that pick parts off moving belts, and at what fast seeing and reacting could mean for home robots and factory cells.

Two players throw their hands in a rock paper scissors contest while friends crowd around. Photo: Jeff Eaton / Wikimedia Commons (CC BY-SA 2.0)

From Han dynasty hand games to the janken robot

People played the game for centuries before anyone thought of building a machine for it. Around 1600 the Ming dynasty writer Xie Zhaozhe described a hand game called shoushiling and said it dated back to the Han dynasty. Hand games built on three gestures that beat each other in a circle travelled from China to Japan, where they were known as sansukumi-ken. One popular version, kitsune-ken, set a fox against a village head and a hunter. The rock, paper and scissors signs of jan-ken took their modern form in the late 19th century, and by the early 20th century the game had reached the West through contact with Japan.

Children in Myanmar playing rock paper scissors in a circle on the ground. Photo: Mosmas / Wikimedia Commons (CC BY-SA 4.0)

Mostly the game settled small arguments, though now and then it decided big ones. In 2005 the Japanese businessman Takashi Hashiyama could not choose between Christie's and Sotheby's to sell his company's Impressionist paintings, so he asked them to play for it. Christie's took advice from the 11 year old twin daughters of one of its directors, who said scissors because everybody expects rock. Sotheby's threw paper and lost. Computers started playing too. In 1999 a program called Iocaine Powder won the First International RoShamBo Programming Competition by looking for patterns in the opponent's earlier moves and predicting the next one.

A match at the 4th UK Rock Paper Scissors Championships in 2010. Photo: James Bamber via Wikimedia Commons (public domain)

Prediction does little against a person who plays at random, so the Ishikawa Watanabe Laboratory at the University of Tokyo went another way. In 2012 it showed a janken robot with a 100% win rate. A high-speed vision system found the position and shape of the human hand in 1 ms, and a fast robot hand then formed the winning shape. According to the lab, the robot's final shape in the first version came about 20 ms after the human hand had finished. A second version closed that gap almost completely, and a third added fast pan and tilt tracking so the player's hand could move around freely.

The Salisbury Hand of about 1982, an early computer controlled robot hand with force sensors in each finger, at the MIT Museum. Photo: Daderot / Wikimedia Commons (CC0)

Why fast vision matters more than the game

The game still makes news. On 30 September 2026 the organisers of the London Marathon used it to decide which elite field would race on which day of the first two-day London Marathon, in April 2027. Men's world record holder Sabastian Sawe played the Swiss wheelchair racer Catherine Debrunner. Debrunner won, so the women's elite and wheelchair races keep Sunday 25 April and the men's elite field runs on Saturday 24 April. The organisers called it an entertaining way to settle the question. Nobody brought a robot, which is just as well, since the Tokyo hand would have made the result a foregone conclusion.

Rock paper scissors takes the place of a coin toss at a sports day for US Navy and Royal Brunei Armed Forces sailors in 2012. Photo: U.S. Navy via Wikimedia Commons (public domain)

The Tokyo lab presented its janken robot as an example of a person and a machine cooperating within a few milliseconds. It said the same technique could help with motion support for people and let humans and robots work together without delay. The lab also argued that robots faster than human eyes and hands could change the economics of labour intensive work that factories now send abroad. Industrial vision has been moving in that direction. Nikon's robot vision system, announced in 2024, measures parts in 2D and 3D at up to 250 frames per second so that a robot can follow parts that move.

A Schunk SVH servo electric five finger robot hand at Hannover Messe 2016. Photo: NearEMPTiness / Wikimedia Commons (CC BY-SA 4.0)

ABB robots already rely on cameras to deal with moving targets. The IRB 360 FlexPicker, run with PickMaster software, finds items on a running belt and picks them, and conveyor tracking lets an arm follow a part as it travels. Outside software can also steer ABB arms through Externally Guided Motion. That is how Google's GoalsEye research drove an IRB 120T to play table tennis, a project covered in our table tennis robot post. ABB does not make a rock paper scissors robot, but a dual arm YuMi IRB 14000 has played tic-tac-toe against visitors at the FIRST Championship in Detroit.

An ABB YuMi set up to play tic-tac-toe at the FIRST Championship in Detroit. Photo: Stilfehler / Wikimedia Commons (CC BY-SA 4.0)

Where fast reacting robots go next

A home robot that won every round would be a poor playmate, and children would give up after a few throws. The part worth keeping is the reading of hands. A robot that recognises a gesture in a few milliseconds could take a cup from someone's hand, or pause when a child reaches into its path. Collaborative arms such as ABB's GoFa CRB 15000 already sense contact through their joints and stop, and ABB's SafeMove software limits speeds and zones around the robot. With faster vision, robots like these could slow down while a hand is still on its way toward them.

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

In factories the lab's argument about speed still holds. Picking, sorting and inspection lines depend on how quickly a camera can find each item and pass its position to the robot, so faster vision can mean more picks per minute from the same arm. It has costs. Cameras that run at hundreds or thousands of frames per second need strong lighting, fast processors and careful calibration between camera and robot. A robot that reacts in a few milliseconds also has to pass a safety review before anyone works next to it, however good its eyes are.

ABB IRB 360 FlexPicker delta robots picking items on a moving line. Photo: ABB

Skills are the third problem. Building a cell where a camera and an arm work together in tight timing takes optics, control and robot programming at once, and few people learn all three. Students can start in RobotStudio, simulating the arm and testing how a program reacts to positions sent from outside, and then move to a real camera and robot. The Tokyo janken hand is still a lab demonstration more than ten years on. Pieces of it, a fast camera feeding positions to a fast arm, already sit inside ordinary picking and sorting cells.

Students program an ABB IRB 120, the arm family Google steered for its GoalsEye table tennis research. Photo: Peterantala / Wikimedia Commons (CC BY-SA 3.0)

Innovation

AI solutions for effortless ABB robot control.

Automation

Robotics

ceojohntran@chatwithrobot.net

+84905311611

© 2025. All rights reserved.

qtran1215@gmail.com