The students at the Department of Technological Studies at Ohio Northern University are using techniques similar to industrial automatic dynamic bin picking applications (grabbing parts from moving bins) to make a KUKA KR3 industrial robot play air hockey.
They have the arm mounted in an inverted position on a gantry over an air hockey table with a Point Grey Flea2 camera which snaps 30 images per second of the entire table. These images are transmitted to a computer where image distortions are removed and VisionPro machine vision software distinguishes the puck and transfers its x,y position to another algorithm on the computer. This program keeps track of current and previous positions and attempts to calculate the puck’s trajectory and determine where the arm should move to strike it.
Since the arm can only move at up two 2 meters per second, which is much slower than a human player, they have devised a “puck striker” which the robot holds. The striker is like a regular air hockey paddle split in to four quadrants which can rapidly expand outwards at 16 times per second to strike the puck more effectively than the arm alone could accomplish. They designed the striker in CATIA V5 and then printed it on a rapid prototype 3D printer.
They also found that the standard side walls of an air hockey table are constructed from MDF (medium density fiberboard) which has some flex and deforms slightly when hit by the puck. This causes the the puck to bounce off the walls at a different angle than they could accurately predict with a basic “angle of incidence equals angle of deflection” algorithm. To solve this problem they replaced the side rails of the table with aluminum which deflects the puck in a more predictable manner.
When the player scores, the puck is routed out the bottom of the robot’s goal onto a small conveyor belt which carries it under the table back to the human to put back into play. It looks like the robot currently plays mainly defense, but the students hope to continue developing the software with more AI to play more effectively. Then it should be ready to fill in and help satisfy your late night air hockey needs after your roommate passes out. After that, maybe their next project will be a robot that can play quarters.
Here is a short video of a student presentation of the project. Unfortunately it doesn’t show much actual hockey playing by the robot, which was still awaiting more advanced software at the time of the presentation.
[Link] to project home page with more technical details and images.
Apparently an engineering services company called Nuvation has made a very successful version of an air hockey playing robot that is able to beat all comers (or at least wins 90% of games). Here is a nice video interview with the CEO of Nuvation explaining their system. One interesting point is that the robot vision system not only tracks the current position, speed and direction of the puck, but then makes intelligent guesses on its future position predicting up to three bounces.
The robot, by the way, does live up to its advance billing, according to Kanellos. In watching it for a half an hour at the Freescale Technology Forum, the bot never lost a goal and scored many. Granted, none of the players he witnessed were off-duty roofing contractors tanked up on a 12-pack of Keystone Lager, but some players were pretty good nonetheless. The most eerie moments come when the robot would block a shot without even moving. It mocks you with its stillness.