Rebound Fumble: Aim Low
January 1, 2013 • ∞
This won’t be good enough. It wasn’t good enough today. It won’t be good enough against anybody else, either. - Bill Belichick
(Please see the introductory post if you have not read it yet)
We know intuitively that building robots is hard. In our opinion, having put together enough resources to field a basic kit bot with a default control system is an effort far and above that typically required of high school students. But it seems like every year at kick off almost everyone vastly over-predicts how good robots will be at scoring points. So we took a look at the 2012 FRCFMS twitter feed data for qualifying matches. We took out MAR, MSC, and CMP because these over-represent robots that score lots of points relative to the field. Also missing are SDC, OR, GT, SAC, UT, CAF, and NV because they didn’t live update to twitter. They are reasonably well distributed throughout the weeks, and they represent around 10% of events, so we don’t think it undermines our main points.
We realize this is fairly long and has big images, but we’ve highlighted the main points. A few tidbits to entice you to read on: Did you know that the median alliance scored only 4 points in teleop? Or that 10% of qualifying alliances didn’t score a single point in 2012? Let’s begin.
#1: Alliances don’t score many points. The Median Alliance scored 17 points. The Mean alliance scored 19 points.

The Median is the 50% mark. That is in 50% of matches teams scored more points than that, and in 50% of matches teams scored fewer points. The Mean is less resistant to outliers. For example a score of 1 MILLION points by one alliance all season long would not change the median since it it is one data point out of about 10,000. It would however make the mean score 121 points. You can see that about 10% of the time an alliance fails to score a single point – thanks to fouls this is actually a lower number than you see historically (we’ll explore this in a later post).
We realize that by individual data point it is hard to look at, so here’s a histogram where each bar is 3 points.

The big picture is obviously important, but what about the individual phases of the game? Since the FMS twitter feed breaks down the scores by hybrid/teleop/bridge/fouls we’ll look at each piece to try and understand how that affects the final outcome of a match. Let’s start with Hybrid mode.
#2: In Hybrid, the Median Alliance scored 6 points. Less than 5% of Alliances scored more than 18 points! (more than 3 balls in the top goal)

The neat thing about this method of presenting the columns is each column is one ball scored in the top goal. You might think of it as a “scoring action.” So you can count out how many balls you thought the “typical” alliance might score and see where it fell on the real continuum. One thing to note is that even though the GDC often offers easier goals, teams don’t often take them up on it. They mostly shoot for the 3 pointer. Once Hybrid is done, the next logical place to look is teleoperated mode when the driver’s take control.
#3: In Teleop, the Median Alliance scored 4 points. Only 25% of Alliances scored more than 9 points (3 balls in the top goal)

Again, think of each bar as a “scoring action” and remember this is per alliance, not per robot! If you think you end up on alliances with three equal team mates, then you only need to score 1/3 of these points to be “typical”. This was the most surprising result to us. We would have guessed teams scored more in teleop than they did in autonomous. And as teleop draws to a close, teams make for the bridges to balance for regular points and coop points. How’d they do?
#4: On the Bridge, the mean Alliance scored 5, and roughly 10% of Alliances double balanced.

#5: Teams successfully balanced the coop bridge 30% of the time. However, this means that roughly 70% of the time teams left a free “match win” on the table.

While I think (hope?) we won’t see the coop mechanic again, it looks like it was way undervalued. With a mechanical brake it seems like teams really could’ve played havoc with the rankings and the 1 point for an unbalanced bridge.
And finally, penalties play a part in every FRC game. How well did teams play by the rules?
#6: ~75% of Alliances didn’t get Fouls! Less than 15% of alliances had 1 foul. However the average alliance scored so few points, even one Foul was a big deal!
The graph wasn’t that informative for this one, so I didn’t include it. 9+ penalty points happened about 5% of the time, so teams were pretty good about not getting technical fouls or stacking foul upon foul.
The moral of the story is that teams just don’t score many points. In 2012 two kitbots with decent drivers could have outscored 50% of alliances by playing defense and then balancing! Teams think they need to score lots of points, and waste a ton of time designing complex mechanisms that are hard to tune. We think these data point out that designing these complicated mechanisms is unnecessary. Instead aim low and actually execute. Imagine if you had scored 2 balls in autonomous in the middle goal, and then 2 balls in the middle goal in teleop, and then balanced. That would mean your robot, scored more points in a match than 70% of FRC Alliances! And with a robot like that, you are definitely going to be in the hunt.
As disembodied Dave Lavery/Blair Hundertmark says every year: “Good Luck, and we’ll see you at the competition!
And here is a bonus pie chart of where points were scored in 2012!
