Air Hockey Robot: Brunswick Tables to ABB Speed

An air hockey robot gets very little time to think. A hard shot crosses the table in a fraction of a second and comes off the rails at a different angle every time, so the robot has to spot the puck, work out its path and get the mallet there first. This post covers how air hockey grew out of physics lab air tables and a Brunswick engineering project around 1970, and how researchers and students turned the game into a test of fast vision and control. Then it gets to Puck Pilot, the UBC robot that beat a former world champion in 2026, and to where ABB robots fit.

Joan Garcia, winner of the Catalan air hockey league in 2003, plays a match in 2005. Photo: Cristina Ruiz / Wikimedia Commons (CC BY 3.0)

From floating boxes to Brunswick's air hockey table

Air hockey started with a conveyor idea. Air tables of the 1940s had holes plugged by ball bearings, so a heavy box could slide on a cushion of air. In the early 1960s John Stull, a professor at Alfred University, built a low friction air track for physics teaching by putting a vacuum cleaner under a drilled surface. The Ealing Corporation developed the idea into tables with a grid of small holes and low pressure air underneath. Light plastic pucks floated over them, which made the tables handy for showing collisions in physics classes.

A college ice hockey faceoff, the game that Brunswick's engineers simplified into air hockey. Photo: U.S. Air Force via Wikimedia Commons (public domain)

In 1969 three engineers at Brunswick Billiards, Phil Crossman, Bob Kenrick and Brad Baldwin, started work on a game for a low friction surface. The project stalled until Bob Lemieux revived it as a stripped down version of ice hockey, with a thin disc, two strikers and slot goals fitted with photodetectors. Brunswick had air hockey on sale by 1972, and it made money from the start. Players in Houston had formed an air hockey association by 1973, and the United States Air-Table Hockey Association, founded in 1975 by Phil Arnold, has sanctioned national or world championships every year since.

A row of coin operated air hockey tables in an entertainment centre in Tbilisi, Georgia, 2018. Photo: Tiia Monto / Wikimedia Commons (CC BY-SA 3.0)

Robot researchers took to the game because it is fast but contained. The puck stays flat on the table and rebounds follow fairly simple physics, yet the speed leaves almost no margin for error. In the early 2000s Darrin Bentivegna and colleagues at ATR in Kyoto taught a humanoid robot to play by learning from human demonstrations. University teams and hobby builders went on to make gantry robots, in which an overhead camera tracks the puck, software predicts its path, and motors drive the mallet along two axes to block or strike.

A mallet and puck on an air hockey table, the small fast target a robot camera has to track. Photo: Stuart Kerr / Wikimedia Commons (CC BY 2.0)

Puck Pilot and the AI air hockey robots of 2026

The UBC robot that made news in September 2026 is called Puck Pilot. Engineering Physics students and researchers at the University of British Columbia built it as a robotics and machine learning project, with the stated aim of beating humans at air hockey. Cameras and reflective markers track the puck, and the system processes its position about 130 times a second, which lets it predict shots and answer them almost at once. When two former world champions from Texas came to play, Puck Pilot beat Tim Weissman 7 to 2. His son Jacob Weissman won his match 7 to 5.

A mallet on a neon lit arcade air hockey table. Photo: Luke via Wikimedia Commons (public domain)

Tim Weissman called the robot "scary good", and the Vancouver Sun headlined the players comparing it to a brick wall. Jacob Weissman said he played a much more upbeat game than usual, which made him harder for the machine to predict. Earlier, in June 2026, another trio of UBC students showed an AI air hockey table that learned entirely in simulation. They built a detailed digital twin of the table, trained the AI there and then copied it onto the real robot, which gave human players a real challenge without a single practice session on the physical table.

A boy plays air hockey, the kind of opponent a home robot table would face. Photo: Jay Clark / Wikimedia Commons (CC BY 2.0)

ABB has not built an air hockey robot, though the problem resembles work its robots already do. An IRB 360 FlexPicker delta robot picks items from a moving conveyor at high speed, guided by cameras and PickMaster software that predicts where each item will be when the gripper arrives. That track, predict and intercept loop is explained in our page on conveyor tracking. A FlexPicker mounted over a table could in principle play the game, although nobody has reported trying it. A team would test such a cell in RobotStudio before risking real hardware.

An ABB FlexPicker cell set up with PickMaster Lite, software that guides fast picking with cameras. Photo: ABB

Where air hockey robots and fast vision go next

For homes and arcades, the next step is probably a table with a built in robot opponent: an overhead camera, a simple two axis drive and software with difficulty levels, sold as an entertainment machine. The UBC simulation project helps here, because training in a digital twin costs less than wearing out a real table and lets a maker tune the opponent before shipping. Cost is the main obstacle. Fast motors, a good camera and enough computing to run at more than 100 frames a second add up quickly for a game table.

A home air hockey table with puck and mallets, the likely starting point for a consumer robot opponent. Photo: N734LQ / Wikimedia Commons (public domain)

Factories may get more out of this research than arcades. A robot that can catch a puck coming off the rails can also grab parts that tumble on a conveyor or recover a part that slips in its gripper. ABB's SWIFTI CRB 1100 is a collaborative robot built to move quickly in shared spaces, and software for predicting motion from camera data keeps getting better. Table tennis research went the same way, as described in our post on table tennis robots, where work on a game fed into fast manipulation in general.

ABB IRB 360 FlexPicker robots picking chocolate balls from a moving belt at Delicato. Photo: ABB

Safety and fairness are still open. A mallet driven by a motor at full speed can hurt a hand that reaches across the table, so a home robot needs speed limits and sensors that stop it when a person gets too close. Players also want a rival they can beat now and then. Jacob Weissman's win shows that an unusual, unpredictable style still works against a learned model, so future robots will probably need to adapt during a match, the way human players read an opponent after a few goals.

An engineer programs an ABB SWIFTI CRB 1100 cobot at a workshop bench, an arm built to move fast near people. Photo: ABB

Innovation

AI solutions for effortless ABB robot control.

Automation

Robotics

ceojohntran@chatwithrobot.net

+84905311611

© 2025. All rights reserved.

qtran1215@gmail.com