Kickoff Strategy Discussions

#1

How do continuously successful teams handle their initial strategy discussions? How do teams like 254 and 2056 continue to determine the optimal strategy year after year? Is there a known or established process? How could a team looking to improve (like mine) improve or refactor our early build season discussions? What is your process?

4 Likes
#2

Look at things like the strategic design seminar that was just given at Detroit, and other resources online. They are generally very good.

4 Likes
#3

First we all spend some time after the kickoff broadcast to read the game manual. We bring printers to our kickoff event and churn out a few copies, if some students or mentors prefer to mark up paper copies as opposed to reading the PDF. Everyone reads over the game and rules chapters. The appropriate subteams will eventually go over the robot construction and wiring rules sections.

The afternoon of kickoff we brainstorm and list all of the possible actions and point values in the given game. We then try to work out what are possible “plays” - combinations of actions - and what are typical cycles going to look like. We begin to get a sense of what “core cycles” might be key to the game. For example in Deep Space, basic core cycles are “grab a hatch panel from the feeder station and place it on the cargo ship or rocket”, grab a cargo from the floor and place it, etc. We sometimes find supporting plays like in Stronghold “grab ball from center field, get it through a defense and pass it to a high goal scorer”.

On Sunday after kickoff we try to have a sense of what the key core cycles could be, how long we expect them to take, how many points they’re worth, and what is the difficulty of executing the gameplay (the high goal in Steamworks has a much higher chance you’ll miss the shot). This includes an analysis for raw points as well as for RPs. By the end of the day we have a sense of what the most valuable actions are going to be for the game. We come up with a priority list of things we feel we MUST do, what we WANT to do, and possibly what we know we DON’T want to do. The team votes on all this and a consensus is reached.

Throughout the strategic analysis we concentrate on “what” without really worrying about “how”. However we do keep in the back of our minds that our fabrication resources are limited (finding a sponsor-provided build space is a perennial issue on our team, and we are also working with basic shop tools only). The KISS principle is referred to frequently. Whatever we choose to do has to be doable in a simple way if we expect the robot to be done on time.

Steamworks was a clear example. The math said that gear cycles were a necessity and the climb was very desirable, but that the fuel cycles were risky, difficult and not really worth much. So the decision was to make a robot that did gears and climb, but didn’t do fuel.

Once we have these decisions we flesh out a functional specification for the robot, a representative from each subteam signs off (basically to prevent conversations that start with “but I thought we said that…”) This drives the prototyping stages that come next.

The biggest advice I can offer to any team (including my own) is to be willing to put your foot down on what NOT to try. This year the math said to focus on having a level 3 hab climb and level 1 hatches and cargo. So, those were the top of our priority list. We could have just stopped there. Instead we said “and we also want to do level 2 and 3 rocket, although at a lower priority" on the list. In other words our priority list said we wanted to try to do everything. If you scouted our robot, you’ll notice that we do hatches and cargo on all 3 rocket levels, but we had no climber at our district events, and bolted on a level 2 climb at DCMP. In other words, our priorities shifted sometime during build season.

Although we did quite decently this season, I think the outcome could have been different if we had said “level 1 game pieces and hab 3 climb, full stop” and built to that spec.

6 Likes
#4

Karthik just released this , I think just about the whole video is interesting, but there are a lot of important things in there too. 1678 releases strategic design videos that cover ALL the basics at least design wise. This video is also great it explains some smaller details in the process. there are tons of other videos out there that 971,1678, 1114, and 254. you just need to look. also try to get in contact with some teams around you or afar and see if you can discord call or even meet in person to talk about it.

7 Likes
#5

As mentioned above, there was a strategic design seminar at Detroit. There was also a seminar done at Capital City Classic by Mike Corsetto, mentor of 1678. He covers some of this process. I have where he starts talking about it time stamped in this link.

https://youtu.be/dSXDV-UhhxU?t=2008

1 Like
#6

Another question. How much time do you spend on each of these elements?

Like reading the game manual. That can take an entire day to read through the whole thing properly sometimes.

In the past we’ve split into groups and each group read a section then tried to give everyone else a brief update on their section. But I find they can sometimes miss key elements. And we can spend an hour (or more if necessary) on it. But a lot of our students can get distracted easily and it starts to become inefficient.

1 Like
#7

Our team totally called the requirements, except for the necessity of multiclimb to level 3.

You need to remember that 254 made major changes between their last regional and Worlds.
I think that the biggest point here is not to get lost in your initial design. You have to use a process of continuous improvement.

I know the biggest difference between us and them is packaging. They package real tight, where we over engineer. If we had been packaged better, we could have added the suction climber that we designed for worlds. The weight just wasn’t there for the change.

#8

We break into smaller groups, but every group reads the entire section. Yes, not everyone is engaged, but we spend enough time talking up how important this is that most kids get it. I think we allot 2 hours for rules reading before we gather back together to continue the game analysis.

The first weekend is long and sometimes tedious. What I described in my post above takes us all weekend. Full 9-5 days both Saturday and Sunday.

1 Like
#9

Keeping attention while reading a 100+ page manual is tough for adults and teenagers. The best process we’ve found is similar to GreyingJay’s: We break into small groups of no more than 5, each with a mentor or senior student. Each group does an oral read of the whole manual (or at least the game sections if they’re slow) for about 2 hours. The oral read works like this:

  • A student will read one paragraph out loud so that the others can hear it.
  • That student will describe what they read in their own words (or ask for help if they didn’t understand).
  • The rest of the group will decide if they agree with the interpretation, and ask each other questions about the section if they have any. If we have an unanswered question, we write it in a list to be revisited later.
  • When we’re all good, we move on to the next paragraph, and the next student’s turn to read aloud.

When we get back to the larger group, we go through a standard game survey we’ve built over the years (like how many game tasks, how many gamepieces, what kind of defense rules etc). We also review everyone’s list of unanswered questions, and if none of us can answer them, we mark it as something to ask other teams or in the official Q&A.

Having team members rephrase each rule in their own words is particularly good at building understanding I find. It’s a bit slow, but you have to be slow when digesting something so complex as a new FRC game.

4 Likes
#10

This is key. Unless you read the game manual (at least up to the start of the the power section - R39 this year) you cannot really talk about the game and start thinking about what you want to do. You have to read up to the end of the Game rules at an absolute minimum, but the lower robot rules give you an idea of the flavour of the challenge (size constraints, starting configs, weird bumper rules, etc).

While you want to separate strategic analysis from the mechanism you would use to solve it, it is important to understand stuff like no shooters allowed/restricted starting height/high goals, which would mean an extendable lift/reach is required. This is because the end of the strategic exercise is going to be defining what you want functionally wise. You then need to “cost” your desired functions from a complexity point of view and ensure they are within your limits. We tend to use degrees of freedom, and number of separate mechanisms as a first pass at this.

The other part of game analysis is the scoring. What are the max scores for each type, points/minute, cycle times, beginning, middle and end game trade offs.

Our goal is to get to this point by Sunday night of kickoff weekend. We do game manual read, discussion, cycle definition on the Saturday. Do the scoring analysis as an offline exercise, and then come together with a smaller team Sunday morning to do a cycle timing analysis. Sunday afternoon is a full team discussion about what core cycles we are going to focus on and hence what the initial spec is.

Once you know the scoring rates, the cycles, and the complexity costs it becomes relatively easy to maximize the scoring at whatever you set as your maximum complexity. That drives a preliminary spec and drives prototyping. After you validate that you have mechanism solutions or not, reiterate the spec until you do, freeze it, and start designing.

As @GreyingJay said, it is a REALLY good idea to set a focused and doable spec. Especially next year with no bag, you can always iterate. The key take away is that it is better to do a very limited number of things really well, than to kinda sort do everything. 100% repeatable is what you should be aiming for. Last year, we aimed too high and it showed.

Finishing early, getting drive time in and showing up at your event ready to compete (and maximizing your time practicing on the real field) is likely the most deterministic factor in how well you will do.

2 Likes
#11

Yes! The team I mentored this year broke up into 4 groups, each reading a different section and then described what they found to the rest of the team. Due to lack of experience, they spent a significant fraction of their time on parts of the Game Manual that would not affect strategic decisions i.e. what colour the wires need to be.

Based on the really basic questions I was getting throughout the build season, it became clear that the student team members only had a very superficial understanding of the Game Manual. I doubt that any of them read through the whole Game Manual. As a result of this superficial understanding of the game, we missed the strategic importance of the Null Hatches and built a moderately effective Hatch bot and never completed the complex ground pickup Cargo mechanism. In hindsight, we could have spent a lot less effort to build a much more effective Cargo bot without ground pickup.

#12

In addition to Karthik’s presentations, my go-to is reading JVN’s design process for robotics

Scroll to the bottom and check that out in addition to the weighted objectives table.

1 Like
#13

I think you’re looking at this from the wrong end…

It’s not hard to figure out the optimal strategy each year, it’s always “score more points than anyone else”. This year, you could be more specific: 12+ cargo/hatch cycles, and a sub 5 second lvl 3 climb. It’s easy to figure out that’s a winning strategy.

The hard part is figuring out your team’s ability. Can your team actually design, build, and execute that optimal strategy? For >95% of FRC teams, the answer is no.

Borrowing from Karthik’s presentation, many of the perennial powerhouses might have 100 robot points to split among ten tasks, for ten 10/10 mechanisms. Most FRC teams might have something like 50 robot points to split among those ten tasks.

The difficulty is figuring out where you can optimally spend those 50 points. It’s not hard to spend 100 robot points, just do everything perfectly. It’s much harder to identify the 5 tasks that at 50-point team can do perfectly to get the best result. It’s hard to differentiate between the worthwhile and the distractions.

If you want to match 2056’s consistent performance, you need to build your team’s resources (people, effort, money, and equipment) so that you’re able to do everything perfectly, every year. If you feel that you’ve already maxed your team’s resources, or you aren’t at the top tier of resourced teams, you need to optimize how you spend your resources.

Don’t do everything, chances are you won’t be able to. You’ll be much better served at doing half the things really, really well.

As for what those things are, it probably depends on what your team is historically good at doing, what events you’re competing at (early events, late events, who’s at them?), and a bit about the game design.

You’ll want to prioritize different game tasks depending on if “your team improving” is defined as winning a regional, being competitive at CMP, making it to DCMP, playing in playoffs at one event, etc. The answer will be specific to each team and their situation.

5 Likes
#14

I think something I’ve been reminded of this year is that “mechanism” doesn’t always refer to game-specific, point-scoring parts of your robot.

In 2017 we tried to adopt a philosophy of “if it moves… we measure it” based on the same philosophy getting another team an Innovation in Controls award. We put limit switches and sensors on everything – including things that really didn’t need to have them – and then we spent too much time trying to get our software to deal with all that information. Adding sensors requires mounts and wiring and software and problem solving time. Adding multiple sensors exponentially increases the combinations of scenarios and errors you need to deal with (what do you do if one of them fails) Some of it was definitely useful but some of it was eventually declared superfluous or too inconsistent to be useful and ultimately ignored. Overly complex sensors are the equivalent of building another physical mechanism.

We always try to get “bling” (RGB addressable LED strips) on our robot too. Students love seeing the pretty lights. But this too is a mechanism.

So not only are we trying to build intakes and shooters and elevators and climbers, but we’re also trying to build fancy sensors for them and we’re also trying to pretty up the robot with LED lighting. Those don’t come for free, and need to be accounted for in your resources. This year our students literally had to decide: do we want to finish installing and wiring the LED strip lights, or do we want to bolt on a climber? We only had the resources to do one.

2 Likes
#15

I’m assuming you made the correct choice and installed the LED’s? :stuck_out_tongue:

4 Likes
#16

We are definitely not in that top 5%. But we have a plan to get there. We know how to do certain things well. Where it would take some teams 10 pts to implement their lift, it only takes us 3 pts. That conversion makes our points go farther. So each year we try to add something new to our list of specialty skills. In the off season we get all of our newbies up to speed on the team’s specialty skills. If we put enough things into that category, eventually we will be competitive with the best teams.

#17

This the the key and hardest part of the process. First you have to understand the game and the cycles, then you need to choose the highest value cycle that your team can deliver. Assessing the maximum complexity you can deliver is tough. The key thing is to spend your complexity points such that you do the tasks you choose at 100% success, not 75-80%.

1 Like
#18

The following is the kind of analysis some teams do, and I wish more did. It doesn’t really have much to do with ‘cycle time analysis’ or ‘optimizing pathing’ or ‘ideal alliance compositions’, so it’s not very flashy. But given that the majority of FRC games boil down to “do one cycle as many times as you can in 2:30”, this is probably more than good enough. Things get a bit weirder if you’re in the top or bottom 10% of teams, but chances are – especially if you’re reading this – that you’re not. This is really simplified, but someone once told me KISS. Don’t @ me.


Let’s do some math:

At the beginning of the FRC 2020 season I recognize that there is one game piece and a low and high scoring location. Low is 1pt, high is 2pts (remind me: January 4, 2020). I break this down into 5 main robot “features”. Each has a difficulty of 10 robot pts, and I’ve FAIRLY evaluated my team as having 40 robot pts. The five “mechanisms” are:

  1. Drive
  2. Pick up game piece
  3. Elevate game piece
  4. Align to scoring location
  5. Score game piece

I identify two paths my team can take:

Score high+low: I do all five tasks at an advanced level, but I leave out reliability. My mechanisms are solid from design, build, and execution standpoint, but every once in a while they have a “fluke” and don’t work. (raise your hand if a team has ever told you: “oh in that match our elevator/arm/whatever was homed wrong, but we fixed it, and it’s not going to happen again”) My points are assigned like this:

  1. 8/10
  2. 8/10
  3. 8/10
  4. 8/10
  5. 8/10

Score only low: I give up only the elevate task, but make a perfect mechanism for the other for tasks. I’ve used the time I save from not making an elevator to design out all the “mistaken homing” situations that could occur. Here’s my robot point breakdown:

  1. 10/10
  2. 10/10
  3. 0/10
  4. 10/10
  5. 10/10

Bit of a jump here, but let’s assume pts/max is the chance of success because even though real life is hard, math should be easy.

If both robots can do ten cycles in a match, and low goals are worth half as much as high goals, which robot should I build? The 80% robot that scores high or the 100% robot that scores low?

Don't you know the answer without clicking?

Each of the 5 (or 4) actions has to be successful to have a successful cycle.

Each cycle of high bot has five actions, each with a 80% chance of success. Each cycle is worth 2 pts, and I have 10 attempts.

0.80^5 \times 10 \times 2 = 6.5 \text{pts}

Each cycle of low bot has four actions, with 100% chance of success. Each cycle is worth 1 pt, and I have 10 attempts.

1.00^4 \times 10 \times 1 = 10 \text{pts}

Low bot is less flashy, but averages 3.5 more points than the high bot. Any team that you would want to be picked by would recognize that the low bot is worth more points than the flashy high bot, and pick low over high.

3 Likes
#19

Lots of good stuff! As noted before, as you work through your initial strategy, focus on WHAT you want to do, not HOW to do it. About the only reason you should be thinking of mechanisms is to enlighten your estimates of difficulty and costs (for most teams, more time than money).
It’s also a good idea after you get down to your best two or three strategic plans to brainstorm weaknesses/counter-strategies to yours and to re-read the rules with those strategies/counter-strategies in mind. This process will not only help you choose the best strategy, but help you more clearly define your requirements so your strategic capabilities can be more robust. An example might be noting that you need to be able to line up to within 2" of a center line, and be resistant to pushes in THIS direction while scoring.

#20

With the end of bag day I think that there will be a lot of teams that can improve what they would have been stuck with last year. Maybe doing something 50% or 75% next year will be fine because you will have a second build season to improve the robot.