The Paradox of Cycling

Hello,

I’ve been thinking a lot about FIRST strategy recently. Mostly because I’m in the process of writing a whitepaper on the subject.
One subject that I’ve considered at great length is the “Paradox of Cycling”. If you were to run some calculations at the beginning of the season you will see that picking up frisbees (in the case of 2013) is a lot faster than going to the feeder station every time. However in practice a cycling robot often outscores a pickup robot. This phenomenon was decently balanced out in 2013 but I still feel like it was there. Look, we had three cycling robots win world championships over pickup robots, full court shooters, climbers, and combinations of those.

Now I wouldn’t bring up a phenomenon like this without offering a possible explanation.

I think that it has to do with repetition and rhythm for the drivers. I think that a lot of teams design robots to perform a task and completely overlook the fact that if the drivers can’t make the robot perform the task then it doesn’t do any good to build that robot. A lot of teams solve this problem by building practice robots which helps a lot but doesn’t solve the problem completely.

In 2012 no one really had any choice but to pick up off the floor. But if you go back to the videos from finals on Einstein that year you can watch as 180 and 25 sit there awkwardly for a few moments before deciding where to go to pick up balls.

In 2011 it was event worse. A huge majority of teams built robots to pick up tubes off the floor but then they would just drive around among a sea of tubes looking for the shape that they needed. But the very few teams that built cycling robots would just zip back and forth from the feeder station and not have to worry about finding a tube.

It’s the difference between “hmm, where’s a gamepiece…” and “now back for another!”. With a strategy that revolves around cycling it allows drivers to get into a groove and stay there. It is more relaxing in matches at competitions not having to make a decision for every single gamepiece. It also frees up more brain power for the drivers to handle issues in a match (defense, broken robots, ect.).

What are your thoughts on this subject? Does the “KISS” method apply to strategy or is versatility more important?

We certainly didn’t calculate it this way, and I imagine a lot of others didn’t either.

Picking up 4 frisbees means 4 frisbees in randomish locations.

I think if you are going to compare this year to 2012 and 2011 you need to mention how game pieces are entered into the field. In 2012 and 2011 there was little need to cycle as the human players could put the game pieces right next to the scoring area. That was not the case this year and thus the rise of cycling.

Like Adam, all else being equal, we felt cycling would be easier/quicker than picking up from the ground.

I’d look at the game pieces themselves. Manipulating a ball through a robot has been done many times by many robots, with great examples of how to pick them off the floor.

Picking discs off the floor is a unique challenge; consider the fact that the top and bottom have different profiles and flight patterns.

One last thing to consider: The better the competition, the fewer the discs on the floor. While picking up an extra 2 or 4 discs during autonomous gives a team an extra edge (and IMHO is the difference between the best robots and the very good robots), it wasn’t required to win regional competitions, nor is it a requirement to be a very good 3rd pick at a large event.

I don’t think it’s as much of a paradox as you might think.

The primary reason 33 choose to pick up off of the ground was to enable the 7 disk auto. We saw this as an opportunity to control the flow of the match from the beginning. It was also an opportunity to stack the secondary qualifier used for seeding in our favor. For example, if we didn’t pick up off the ground we wouldn’t have seeded second on Archimedes. (It could most defiantly be argued that the reason we lost Einstein was because we started missing the 7 disk.)

The secondary reason was more in-match strategy related. We recognized that there were going to be full court shooters this year, many missed shots, and don’t forget the rain of Frisbees that was changed right at the end of build season. All of these pointed to fast efficient floor pickup as being a valuable asset in addition to cycling. Our Archimedes alliance strategy (double fcs) wouldn’t have possibly worked with us only cycling.

It seemed to turn out okay. We won three districts and were world finalists which is the best we’ve finished in a while (arguably ever.)

, Bryan

I’m just going to post this http://www.thebluealliance.com/match/2013casj_sf1m1 to show how a good floor pick up and a good cycler combo can be used. I agree that having a routine for the drive team is important, but the major problem with great cyclers is that even a super fast robot takes about 4 seconds plus loading and shooting time to shoot four disks. The tradeoff occurs in how fast it takes a drive team to gather and shoot 4 disks from the floor compared to the cross court. The best option to score disks(w/o defense) is to shoot full court.
A good analysis of the cycler vs floor pick up option is the newton top alliance. 1986 and 1538 both had floor pick ups. It was advantageous for 1986 to cycle because they could not pick up the four disks fast enough to match it due to the random disk placement. 1538 would stay and pick up missed disks because they could beat their cycle time picking up off the floor.

We didn’t want to be a cycler, but pickup was just too difficult. So, we just got really good at cycling. If we had gotten pickup working, we’d have done that.

I’d like to see these calculations. We certainly didn’t figure it this way, and I’m with Adam–I doubt it was an overly common assumption, even ignoring the novel difficulty of disc pickup. Certainly it’s possible to setup such calculations to accurately reflect the floor/feeder comparison. Further, despite the similarities in basic description, the strategic usefulness and tradeoff of floor pickup in 2013 is nothing like 2012 or indeed 2011.

In addition to noting the multiple differences in field layout, I’m sure we’re not the only team this year that experimented extensively with getting discs any consistent useful distance from the HP station. Unfortunately (and surely deliberately), this wasn’t possible the way it was in 2012. The corollary is equally important: good HPs (or good robots, e.g. 16) could roughly co-locate quite a few game pieces in 2012. In terms of execution, this is the main bottleneck in floor strategies.

I can think of a couple of reasons that cyclers did well:

  1. There is no weak matchup. There’s no set of opposing robots that is your cryptonite. If you have a FCS, for example, your point output changes dramatically depending on the quality of defense.

  2. Reliability. Cyclers are mechanically simpler than a floor pickup or anything with a climber, and on par with a FCS. So an alliance made of cyclers would be more reliable than anything but an alliance of three full court shooters.

These two weeknesses are compounded by the fact that championship is 5 rounds of eliminations. Playing five rounds of eliminations is very different than qualifications. If you have a robot that scores 100 pts 80% of the time that might not be as good as a robot that scores 50 pts 100% of the time. If you’re the robot that scores 100 pts 80% of the time, it’s probably for one of two reasons:

  1. The robot fairs poorly against certain types of defenders, or
  2. It sometimes has mechanical problems.

In the first case this is bad because it’s correlated with who your opponent is. That means if you lose one against a certain set of opponents you’re more even more likely to lose again than would be expected based on the overall strengths of the alliances. And since you have to win 2 of 3 against each set of opponents, it doesn’t matter how thoroughly you may have crushed your earlier opponents, there’s a matchup you can’t handle you’re done.

The second case, that of mechanical problems is also made worse during eliminations due to the reduced turn-around time between matches. And again it’s worse not only because the matches are close together but because the matches that are close together are likely to be the ones that add to each other. So if a robot breaks enough to be out for two matches the odds are pretty good that those two matches are against the same opponents, which again means that it’s more likely to cause an alliance to get itself eliminated than if there were just two random matches for which it was broken.

It may sound like I don’t like the tournament format, but I do and I probably wouldn’t change it much if it was up to me. It’s just interesting how tournament format changes what kind of robot you want even with the game kept the same.

A couple random points:

-The technical challenge of picking a frisbee up is what made pickup in general a ‘slower’ process. The teams who really nailed a pickup were certainly able to put together a fast cycle if the 4 frisbees they were gathering were in somewhat close proximity to each other.

-1477 was a cycler, but also had a very good floor collector, a strategy we are very comfortable with as well ;). A 7 disc auto, and the ability to ‘steal’ the opposing teams discs were PIVOTAL factors in their alliance persevering.

-Brando

It’s not really a paradox, in my view. It’s just about tradeoffs between different strategies and the designs that are required/effective for executing them.

It is a very effective alliance strategy to first prioritize about depositing all of the 51 discs from the alliance station into the high 3 point goal. If you do this, it is very rare that an opposing alliance will outscore you in teleop. Add a floor loader to the alliance and they can pick up misses, opponents’ misses, and loose discs to swing the teleop score in your favor.

Then the extra auto points (presumably some extra auto disks coming from that floor loader) and extra climb points beyond 10 are simply to raise the alliance’s point ceiling.

It just so happens that the simplest possible robot design that can be a part of an alliance that can execute this strategy is a cycler. So it is not surprising that there are a relatively high number of effective ones in relation to other styles of robots. Since the population size of effective cyclers is larger than any of the other types of effective robot designs, it makes sense that in general the best cyclers will have better accuracy than the best of the other robots, so they are very good at taking discs from the feeder station and getting them into the high goal without risking giving those disk points to the opponents. Remembering that we stated that this was the priority for the alliance strategy, we can begin to see why cyclers emerge as an attractive engineering solution to this year’s game.

Our reasons for floor pick up were almost identical to 33. We spent several hours simulating the game the day of kick-off. After doing this we being able to pick up off the floor gave teams the following advantages:

  • Floor pickup allowed a team to run a 7 disc auto.
  • The 7 disc auto allowed you to stack the auto points in your favor
  • The 7 disc auto gave you a huge advantage in a match. An alliance that scored 13 in auto had 24 point advantage on an alliance that scored 9 in auto. This was the equivalent of 2 additional cycles an alliance would have to score to overcome that deficit.
  • Similar to other projectile games (2006 and 2012) very few teams would be greater than 75% accurate. This mean there would be discs on the floor for us to pick up most of the time.
  • Prior to the rule change, human players could throw discs down field. Having a floor pick up allowed a team to pick up discs that their human player threw down field for them, saving them the trip to the other side of the field and back.

Here’s our experience with having a floor pick up:
-Designing and building an effective floor pickup was very difficult and required a lot of tweaking to get it just right. We were constantly changing little things about the floor pick up through Champs. Arguably the hardest system we’ve had to design in our 9 years.
-We had a 7 disc auto. This was important for different reasons as the season progressed. At our first event (San Diego, week 2) where many teams didn’t even have a 3 disc auto, the 7 disc picked up the slack for them and then some. By Newton when almost everyone had a 3 disc auto, the 7 disc gave us a big advantage over most alliances out of auto.
-Because of the 7 disc we were never lower than 3rd in auto points. At San Diego we were tied for 3rd and at both Inland Empire and on Newton we had the highest auto score of the event.
-Since we were at or close to the highest auto score, we were always at the top of our WLT bracket.
-In San Diego we did a lot of cycling, however as the season went on and we improved the floor pick up more we found it was faster for us to pick up missed shots from our partners and opponents than to go to the feeder station.

An important thing to remember about cycling is its interplay with Defense. Slowing a a small bit can usually cause them to lose 1 cycle. It takes a conisderable amount of defense to slow them down enough to miss another cycle.
For instance, A cycler that consistently does 5 cycles and hangs may require 10 seconds to hang, and therefore has 110 seconds to do 5 cycles with an average cycle time 22 seconds. Holding this robot up with 10 seconds of defense will move it down to 4 cycles. The trick is, to get them below 4 cycles will require an additional 12 to 20 seconds of “hold-up”. While that doesn’t sound like much, it is actually a considerable amount of time for defense to stop a robot.
If you have 3 robots that can consistently do 4 cycles with defense… you can pretty well deplete the frisbee supply, and you are very difficult to defend.
In my opinion, this explains the World Champions a lot. They had 3 excellent cyclers that could consistenly get 6 cycles each if that was their primary job. This meant that if they each did 4 or 5, the supply should be diminished and there is a lot of extra time for the cyclers to play D or otherwise screw up their oppoennts offense. If one of their teamates dropped a cycle, the other two teammates had capacity to take up the slack.

Driver rhythm is extremely important in my opinion. KISS absolutely applies to strategy.

I think at this point most teams understand the concept of strategy before building. Teams are getting better at building within their means and understanding different limitations.

However, I believe most teams don’t take into account match play in their strategy. It’s easy to write down on paper numbers that show how many points a particular strategy should yield. But so often teams take the bigger number and make that their strategy.

Considerable time should be spent taking into account how much “randomness” your strategy may incur. As a drive coach, I always tend to favor the strategy that will allow my drivers to go out and do the same thing every time. This way they can get really really good at it.

For example, if you had a ground pickup, but cycled 90% of the time, how good will your drivers be the one time the coach tells them to pick up the discs off the ground in front of the goal. For most teams that don’t have practice bots and practice fields, they aren’t going to be that good at it simply due to lack of experience doing that task.