October 26, 2010
More Thoughts on Robo-Umps
Perfection is not an attainable goal with any ball-strike detection system. We're going to fall short whether we use human umpires or technology. That simple fact sometimes gets lost when discussing balls and strikes. It's worthwhile to remind ourselves of that up front.
I appreciate some of the questions and comments in the original post about PitchTrax errors in Game 6 of the American League Championship Series. I would like to elaborate a little on what those errors mean for the technology of calling balls and strikes and possibilities for alternative technologies. The question is how close we can get to a perfect ball-strike detection system, at reasonable cost, without too much disruption to the game.
First of all, we should not be too quick to throw umpires out the window. Of course, human umpires are going to be subject to biases and mistakes. I'd like to see a thorough program to understand the nature and causes of those biases and mistakes. This is one of my major areas of ongoing research. If it's possible, and I believe it well could be, I’d like to see a training program to help umpires to improve. Not simply grading them, but actually providing them feedback in a form that is most useful for improvement. Bruce Weber discussed this in As They See ‘Em.
One of the reasons I oppose dismissing human umpires as the best solution a priori is that they don't tend to be subject to the major mistakes that technology is. When technology goes wrong, it tends to go wrong spectacularly, as we see in the instance from this game. Human umpires may be a little bit wrong a lot more often than PitchTrax, but they're rarely off by 5-6 inches for a whole game.
As to other possible technological solutions, there may well be some. I can't think of any that would address the (mostly minor) problems we have with PITCHf/x while not introducing more major problems of their own.
With sensors at home plate, the question is what kind of sensor? One that might be feasible is an optical sensor, i.e., camera. Putting it at the plate would significantly reduce the challenges that exist with PITCHf/x cameras of knowing the location of the camera relative to home plate or some other fixed field reference. However, in order to improve on or aid PITCHf/x accuracy, the alignment of the camera would need to be known and accurate within 1-2 degrees. A camera placed where players are going to step on it and slide into it is bound to encounter trouble maintaining such tight alignment. Moreover, there's no clear location reference for calibration in the field of view of a camera in such a location.
Tracking radars would not fit under the plate, and in fact, with radars as well as cameras, the information provided by a 3-D view and the time information from tracking the whole trajectory are both very valuable. That requires a radar or a camera located farther away from home plate. Having the information about the whole pitch trajectory reduces the error in the first place, not least by helping us identify when it is a pitched ball crossing the plate and not some other object or a baseball going in a different direction. It also helps identify errors and calibration problems when they occur, and even farther down the data consumption path, it aids in remove or correcting spurious data.
There’s a theoretical possibility for TrackMan radar or similar systems to complement the pitch tracking data we currently get from PITCHf/x, though the case for the additional cost is difficult based on the marginal improvement in plate location data alone.
Tango and others have mentioned the FoxTrax hockey puck. It was developed by Sportvision, the same company that does PITCHf/x. They put infrared emitting diodes in the puck and tracked it with infrared sensors placed around the hockey arena. Emitters in the ball (or on the players for that matter) probably don't gain us any resolution or reliability over what we currently have. That’s my understanding based on conversations with the engineers at Sportvision. Even if emitters did provide additional resolution, placing them in the baseball would be a challenge.
We currently do quite well tracking the ball from the mound to the plate with cameras and tracking the batted balls with radar. Both of those technologies can, and hopefully will, be refined. Whether they will ever be good enough to replace umpires in real time, I don't know.