Juggling Robot: From Shannon's Machine to ABB Arms

A juggling robot has to throw an object on a precise arc and then be in the right place when it comes down, so juggling is a hard test of timing and sensing for any machine. This post starts with juggling as a purely human craft and with the juggling machine Claude Shannon built from toy parts. It follows the research labs that taught robots to bounce and toss balls, then a recent attempt to make an industrial style arm juggle. After that come the places where ABB robots fit, and what this work could mean for robots in homes and factories.

Street performer Harvey Juggling keeps three balls in the air at the 2025 Edinburgh Festival Fringe. Photo: Quejaytee / Wikimedia Commons (CC BY 4.0)

From vaudeville stages to Claude Shannon's workshop

For most of history juggling was a stage skill learned through years of practice. Performers toured music halls and vaudeville circuits, and a few of them became stars. A 1927 advertisement in Variety billed the Italian juggler Enrico Rastelli as "The World's Greatest Juggler", and his act rested on hands and eyes he had trained since childhood. None of it was written down as rules a machine could follow. A juggler knew by feel how high to throw and how long to hold each ball, and could recover from a bad toss without thinking about it.

Enrico Rastelli, the Italian juggler billed as the world's greatest, in a vaudeville souvenir book from 1924. Photo: Unknown photographer via Wikimedia Commons (public domain)

Claude Shannon, the engineer who founded information theory, turned some of that feel into mechanics. In the 1970s he built what is generally credited as the first juggling robot, a three ball bounce juggler made from an Erector Set. He named it after the comedian W. C. Fields, who had juggled in vaudeville, and dressed it to match. The machine bounced three steel balls off a drum head and caught them in grooved cups at the top of their arc, where they move slowest, with no sensors or feedback at all. Shannon also worked out a juggling theorem linking flight time, hold time and the number of balls. The machine is now in the MIT Museum.

Claude Shannon's W. C. Fields bounce juggling machine on display at the MIT Museum. Photo: Daderot / Wikimedia Commons (CC0)

University labs then treated juggling as a serious control problem. In 1989 Martin Bühler, P. J. Kindlemann and Daniel Koditschek built a one bar juggler that bounced two objects in a fountain pattern. A later machine from the same group, the Buehgler, kept two ping pong balls going on a single paddle, and its longest run lasted more than four and a half hours. At Georgia Tech, Christopher Atkeson and Stefan Schaal had a five ball bounce juggler working by 1992. Disney Research published work in 2012 on a humanoid that played catch and pass juggled with people, and in 2020 Ploeger and colleagues had a four joint arm juggle two balls for 33 minutes.

A set of beanbag juggling balls, the kind of prop early juggling machines tried to keep in the air. Photo: William Warby / Wikimedia Commons (CC BY 2.0)

Why industrial arms still find juggling hard

The latest attempt came out of a hackathon. At Viam's Fine Motor Skills Hackathon in September 2026, Mandy Meindersma, Jie Xuan Tang and Aika K tried to make a uFactory xArm juggle two balls, and Adafruit wrote up the project on 13 October. They planned to use no camera at all and to throw and catch in the same spot every time, with a funnel fixed to the gripper. Motion planning turned out too slow, so they drove the shoulder and elbow joints directly, added an elbow flick and raised the arm's speed limit. The arm threw balls up to the ceiling, but the team ran out of time before catching worked.

A visitor plays against Omron's adaptive ping pong robot at Hannover Messe 2016. Photo: Bahnfrend / Wikimedia Commons (CC BY-SA 4.0)

Industrial arms are built to follow a programmed path with high repeatability, and their controllers protect the motors from sudden jumps in speed. A throw asks for the opposite: high acceleration, an exact release moment and a known arc. A catch means seeing the ball and moving before it lands. ABB's arms can be steered by outside software through Externally Guided Motion, and Google's GoalsEye table tennis system drove an ABB IRB 120T that way about 240 times a second. We covered that project in our post on the table tennis robot.

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

ABB does not sell a juggling robot, and we found no ABB juggling demo. Its robots do handle moving objects every day, though. The IRB 360 FlexPicker, run with PickMaster software, uses vision to pick items off a running belt, and conveyor tracking lets an arm follow a part while it moves. ABB has also put robots on stage. Its dual arm YuMi IRB 14000 conducted the Lucca Philharmonic Orchestra with Andrea Bocelli in Pisa in 2017, and a YuMi set up to play tic-tac-toe has faced visitors at the FIRST Championship in Detroit. A juggling YuMi is possible, but nobody has announced one.

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)

What juggling robots could teach home and factory machines

Few people will buy a home robot because it juggles. The skills behind juggling are another matter, because a household robot that tidies up has to catch a falling cup, hand things to people and react when something slips. Collaborative arms such as ABB's GoFa CRB 15000 sense contact and stop, and they can be guided by hand, which suits work beside people. Fast throws and catches sit badly with that design, since speed limits near people are what keep cobots safe. A home robot that juggled would have to do it slowly, or inside a space nobody can reach into.

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

In factories, throwing could turn into a real tool. In 2019 researchers from Princeton and Google showed TossingBot, a robot arm that learned to pick up mixed objects and throw them into bins beyond its own reach. A robot that throws parts could sort them faster than one that carries each item across a cell. It would also bring new problems. Dropped or badly thrown parts have to be caught or contained, cameras fast enough to track a flying object add cost, and any cell with parts in the air needs guarding and a safety review before people work near it.

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

Skills may be the hardest part to solve. The hackathon team found that standard motion planning was too slow and that the arm's own safety limits fought every throw, and they only learned that by working on real hardware. Students who want to try ballistic motion on an ABB arm can first model the cell and the throw in RobotStudio, so a mistake costs nothing, and then tune speed and acceleration in RAPID on the real robot. Juggling arms will stay demos for some time, and the control work behind them carries over to any robot that has to deal with objects in flight.

Harvey Juggling balances on a board while juggling for a Fringe crowd in Edinburgh. Photo: Quejaytee / Wikimedia Commons (CC BY 4.0)

Innovation

AI solutions for effortless ABB robot control.

Automation

Robotics

ceojohntran@chatwithrobot.net

+84905311611

© 2025. All rights reserved.

qtran1215@gmail.com