Early Design Process
The first thing that we did (on Sunday after kickoff) was to determine a high-level game strategy, completely independent of any thought about mechanisms or implementation details. From our game simulation and subsequent discussion, we determined that we wanted our robot to have the following capabilities (I am summarizing the results; in actuality these were determined from weighted objective tables on a more continuous reward scale than "Required" or "Desired"):
Requirement - Accurate shooting from any part of the key
Desired - Accurate shooting from beyond the key up to the Coopertition Bridge
Requirement - Able to shoot 3 balls in under 6 seconds (speed and volume are key!)
Requirement - Able to shoot first ball in under 2 seconds from start of Hybrid Mode (we recognized on Day 1 that we wanted to be able to shoot 4+ balls in hybrid)
Desired - Completely, 100% autonomous shooting. We want to be able to have the same accuracy regardless of who is driving, whether or not they are nervous, etc.
After establishing these baseline criteria, we began to talk about big-picture mechanical design. The discussion of pitching wheel vs. catapult vs. other shooting methods was one of the first things we decided. Based on our experience with our 2006 Aim High shooter (which had a shot grouping of less than 2" on average), and our completion of more weighted objective tables, it was decided fairly unanimously that a wheeled shooter would be our "Plan A".
Two questions immediately cropped up:
1) Do we need a turret?
2) Do we need an articulated hood or other launch angle controls, or is wheel speed control enough?
A great discussion ensued, and most of the team (but not all) adopted the belief that because of the protection of the key, a turret was a luxury and not a requirement. The dissenting team members did agree that we should first build a non-turreted shooter, and could probably add a turret (as weight and time allows) at a later time if we are unsatisfied. As it turned out, neither weight nor time allowed, and we were pretty satisfied with the non-turreted shooter.
On the topic of hood/launch angle articulation, we again erred on the side of simplicity. Those of us (mentors and college students) who remembered 2006 knew that shooters were generally highly sensitive to variation in ANYTHING (angle, compression, ball entry, ball type, wheel spacing, vibration, etc). A simple shooter is one that is easier to tune, usually more robust, and less likely to get "out of tune". We again decided to first aim for a non-articulated hood, with the option of adding complexity if we felt it was necessary.
At this point, it was time to go into detailed design and prototyping, and as a team we decided that the shooter system was the single highest priority for the team - the intake, barrier and bridge crossing, balancing, and ball sorting mechanisms were secondary to being able to shoot (only the drive was more important, but this was low risk for us and required no prototyping).
Detailed Design and Prototyping
In choosing a starting point and prototyping configurations, one interesting point of discussion cropped up:
* Those of us who haven't played a lot of basketball insisted that shooting for "nothing but net" was the smartest bet - you have more area to hit on the downward arc of a lob shot!
* Those of us who had a strong basketball background believed that "the backboard is your friend", and that it is the best bet to make shots!
Two things convinced us that the latter group was "right":
1) We found this article:
http://news.ncsu.edu/releases/money-...ball-shooters/
2) We found that with our 2006 Aim High robot, backboard shots were a lot more reliable - the ball had traveled less distance than with a high lob, so angular errors were less significant. That, and the robot put a lot more backspin on the ball than a human shooter could, which seemed to help significantly.
At this point, we enumerated the variables at our disposal in the design of a shooter system:
1. Single or Dual axle shooter
2. Wheel size
3. # of wheels per axle (1 or 2)
4. Wheel tread/material
5. Wheel speed
6. Motor allocation
7. Launch angle
8. Compression
9. Hood/Top axle profile and material
10. Ball entry force/speed
Obviously, there is an enormous number of combinations here. So we decided to start from a point that represented our best intuitive and scientific understanding of what we thought would shoot best (along with options that were favorable from a weight and complexity standpoint, with the logic that a good, simple shooter is decidedly better than a good, complex shooter).
Our initial prototypes were designed as follows:
1. Single axle (simpler/lighter, and we saw plenty of effective single axle shooters in 06 and 09)
2. Fairly small wheels (6" diameter) (smaller wheels are lighter, more compact, and spin up more quickly than larger wheels)
3. 2 wheels on our single axle (we believed that having two contact points helps to center the ball)
4. Wheels were rubber Skyway wheels from a past KOP (They are very light, cheap, and abundant in our shop)
5. Wheel speed was an independent variable tested with a hand drill (initial calculations suggested ~4000 rpm was a good target for mid-key shooting, which was pretty close to what it ended up being)
6. Our initial thought was to devote 2 0673 Fisher Price motors to the shooter, since it let us keep CIMs in the drivetrain while devoting a ton of power to shooter acceleration.
7. Launch angle was another independent variable. Our prototypes were simply adjusted to aim higher or lower as necessary.
8. We initially tried about 1" of ball compression as a guess based on 2006 poof balls. Our later prototypes let us adjust the hood to test different amounts of compression.
9. Initially, we tried a flat hood made out of bent lexan that kept constant compression on the ball.
10. Ball entry for initial prototypes was by hand, so we could test and observe the effect of different entry velocities (but struggled to feed the ball consistently).
We build about 20 different variations of prototypes (most of which were simply adjustments to a single prototype) using mostly wood and hand tools. Very early, it was clear that our 6" wheels, single hood, and ~4000-5000 rpm shooting parameters could hit our distance and arc targets from the front of the key (but weren't consistently getting the necessary distance from the back of the key and beyond). However, we figured that we would get +20% more distance from a finished, well-toleranced shooter.
Our initial prototypes:
http://www.flickr.com/photos/team341...57628843002609
http://www.flickr.com/photos/team341...57628843002609
http://www.flickr.com/photos/team341...57628843002609
http://www.flickr.com/photos/team341...57628843002609
Accuracy and consistency of our initial prototypes were mediocre. We were making 60% of our shots from the key by the end of week 2, which was still quite poor by our expectations (we wanted 90%+). At this point, we completely re-built the prototype shooter to be more robust, and our numbers started climbing higher.
There are too many things we tweaked to enumerate them all, but here are some of them:
* We tried different wheels (but found that roughtop and KOP wheels shredded balls), but kept coming back to Skyways
* We found that hood design was HUGELY important to consistency. Early hoods were very flimsy and would deflect with each shot. We added ribs to attempt to stiffen them, and these helped somewhat, but you could still see the deflection with a naked eye quite easily.
* Compression and wheel spacing between our two shooting wheels were both varied quite a bit. Eventually we settled on ~2" of compression (between the nearest points of our shooter wheels and the hood).
* We also experimented with contact time/angle and found that the longer the ball was in contact with the wheel, the more energy would be transferred. But we were still pretty happy with the distance we were getting with ~45 degrees of wrap.
Lastly, near the end of our prototyping time we saw this video on youtube, and then started to think about "guides" in our hood to help keep the ball centered:
http://www.youtube.com/watch?v=81vOT2KY2ig
We thought that was a great idea, and decided it would be something to try with our first attempt at a "final" mechanism.
First Design Candidate
We all knew from prototyping that tolerances and rigidness were extremely important in our shooter. Our team has a limited number of ways of hitting these design goals. One of them is the benchtop laser engraver/cutter at my work, which can cut most plastics but not metals. We designed a pair of sideplates and two "rails" for our shooter (inspired by 1726) to be cut out of .25" black Delrin.
http://www.flickr.com/photos/team341...57628843002609
The next day I brought the finished parts in to the school and we quickly mounted them to the tower (using that Hex extrusion that comes in the KOP as part of the Kitbot for standoffs). Initial accuracy was pretty good. We took the guide rails and ground them down on a belt sander until the launch angle was
just right, and haven't looked back since. Compression, the hood, and wheel configuration have not changed since week 4 of build.
http://www.flickr.com/photos/team341...57628843002609
The last two weeks of build consisted of a lot of vision system tuning and control system design for good speed control of the shooter wheel. Intuitively, we decided that using a Jaguar speed controller and 200Hz PID loop (the max Jaguar update speed) would give us the best results for our feedback control. Since we knew we were going to be running a 200Hz loop, we wanted accurate, stable speed feedback at least once every 1/200 of a second. Using a banner sensor and timing the duration between pulses (with two pulses - reflective strips - per shooter revolution), we achieved roughly this level of accuracy. The speed loop itself was very quick to write - it is simple PID with a very high proportional gain (pretty close to a "bang-bang" controller). Note that we discovered a need to power our Banner sensor from a 24V solenoid output at this time, since it would sometimes brown out otherwise.
First Competition
Week 1, Hatboro-Horsham District Competition (the very first MAR event)
On Thursday night before the event began, we had an opportunity to go out on the practice field for several rounds. It was immediately obvious that we were overshooting every ball (right over the backboard!) - apparently the balls directly from Gopher (that we were using for testing) are SIGNIFICANTLY softer than the FIRST-logo balls. Luckily, we were able to tune the speed setpoints pretty easily in our SmartDashboard software, and by Friday morning we were ready to go. We went 17-0-1 and won the competition.
Boston Regional
In Week 2 at Chestnut Hill, we duplicated our Week 1 effort and again won the event with a record of 16-2-0. However, at our next event at the Boston Regional, things started to go downhill.
All of sudden, we were undershooting the majority (but not all) of our balls. I tried turning up our speed setpoints, but it wasn't helping. At the same time, we noticed our rate of fire (software controlled - it will not shoot until the wheel is back up to speed) had increased. After a few rounds of terrible shooting accuracy, it dawned on us (it should have been obvious earlier!): we were experiencing wheel slip issues. Our Skyway wheels had been worn by so many shots already that they had become slippery. Luckily our pit neighbors, The Pink Team, had some silicone-based grip tape that they lent us. After applying to our shooter wheels, our accuracy was better than ever, and we went on to win the event (with 233!).
On to the Championship and Beyond
At MAR Championship, our new and improved shooter continued to be a workhorse, and it carried us to the #1 seed and the win for the fourth time of the season. At the World Championship, we likewise shot well for most of the competition. There was one qualification round where our "tower" motor (that runs the elevator that feeds into the shooter) had stalled for several seconds in hybrid mode due to a ball jam, and after that we were undershooting significantly. We swapped out the motor and were good to go.
However, in eliminations, we noticed that we were again not shooting well. I believe that the balls used in eliminations at the Championship felt different than any others we had used all season long, even if they were from the same "batch". We were not the only team to complain of this. Additionally, we found that our silicone tape had started to wear down and was no longer gripping the balls as well as before. I misplaced our timeout ticket (and am still kicking myself for it), so we weren't able to do the repair completely and lost all of the tape in the Finals. We knew our margin for error against a superb 233-987-207 alliance would be slim, and they made us pay!
At IRI, we had replaced the tape and shot well all day, coming in #1 seed and ending up with an OPR 11 points higher than anyone else (!). But once again, a combination of machine problems (mostly with other parts of the robot), bad luck balancing, and unforgiving opponents knocked us out. But we can hang our heads high!
Takeaways and Conclusions
We "guessed right" on many aspects of our shooter this year, which meant it took fewer iterations to get to an acceptable solution than it could have otherwise, and our acceptable solution was very simple and fairly light weight. That was huge. Study of past robots and past FRC games was instrumental to our success.
When prototyping, it is sometimes difficult to figure out if deficiencies of a particular design are limitations to the design, or problems caused by the implementation (in this case a flimsy prototype). Knowing the difference takes experience, targeted experimentation, and, sometimes, building better prototypes.
Lastly, winning (especially at Championships and IRI) has as much to do with "who's left" as it does "who's the best". Designing certain parts of our robot for easier repair would have been very useful. You are VERY rushed during eliminations at both the Championship and IRI!