How to balance an FRC game?

Inspired by the discussion of the Top 5 Favorite FRC Games I am curious how best to balance a game to be competitive for new and old teams.

What makes the low goal in 2016 viable, the low goal in 2022 not viable, how to determine RP boundaries and point scoring. Is it as simple as slapping a set point value for reaching a bonus objective (breaching 2016) or do you make it the same point values as other levels (rocket 2019). What is the best balanced FRC game and why have we or have we not balanced games similarly since?

If anyone has game design experience I am curious how to make these decisions and what makes a well balanced game.

7 Likes

I think there are 3 main factors that contribute to why Low goal worked in 2016 but not as much in 2022.

  1. The high goal in 2016 was a more difficult shot in 2016 than in 2022.
  2. Low bar forced design choices that favored low goal shots. No such game objectives in 2022.
  3. Vision tracking has improved significantly with Limelight. Many teams in 2016 could not target effectively for a high goal shot to make it worth more than low goal.
3 Likes

There were other sources of points in 2016. Driving over the obstacles was worth points, no form of that existed in 2022. Similar thing can be said with 2014 were assists were worth points.

That’s very true. Completing cycles gets you points by itself no matter which goal you scored.

Generally in game design balance is hard. Like. You will 100% screw it up hard. It’s more of an art than a science.

Go do a search to see how many balance adjustments games like Overwatch have had to have. It’s an ongoing thing because it turns out competition among players is a really good optimization function.

You can simulate it and get close this is generally what teams do when they do game analysis. But you bake in certain assumptions.

I guess I’m theory you could toss some sort of ai system at it to see if it could break it or if the optimal game play resulted but that’s not without assumptions around cost functions.

The best tools available are to be able to monitor and adjust - but that works less well in a system like frc where the iteration costs are more real than they are for time spent playing a video game.

(This is based on years of playing frc games as well as some of the stuff they taught me in college years before overwatch or most of the competitive esports scene dropped…)

6 Likes

What if the most balanced game is (a future game) that has frequent adjustments mid-season to keep it balanced.

If we all went into the season knowing that “The point values for low/high goal and RP threshold for cargo/hangar are likely to and allowed to change” there wouldn’t be an argument of “we designed our robot for a specific set of point values” and, assuming good balance decisions, we would all play a more balanced game as it evolved through the season.

3 Likes

Counterpoint: rule changes/clarifications during the season can dramatically shift the difficulty of the task and thus make it significantly overvalued/undervalued. Key examples:

  • 2011 mini bot race: the rulings that allowed motor gearboxes to be removed, the control system to just be a switch, etc. Changed it from a mini robot design challenge to a pure optimization challenge (which tended to favor the teams that could spend endless $ on Tetrix motors…)
  • 2017 (Steamworks) climbers: the ruling that velcro was a rope. The point value was appropriate for the field rope, but made it a must-do because it became so easy with the velcro “ropes”

The problem with changing point values or aspects of the design challenge during the season is that teams establish their designs pretty early in the season based on strategic tradeoffs. It’s frustrating to teams when the strategic choices they made are invalidated. Doing that would very much advantage teams with the resources to iterate quickly as the rules evolve.

7 Likes

The key design element of FIRST games (which is part of the brilliance of FIRST) is to offer enough different scoring opportunities to allow teams across the experience and resource spectrum to execute a robot that offers something to an alliance. So, balance, in my mind should always keep that in mind. In general, FRC games will involve 3 or 4 different “tasks” that each offer a design challenge and that force trade-offs. Games that had good balance rewarded teams that could do all of them reasonably well over teams that were exceptionally good at only one.

Oftentimes, the lack of balance in the game is a result of poorly balancing the scoring value of each task rather than the tasks themselves. A recent example of this was Steamworks where the point value of shooting the fuel into the boiler was very low compared to getting the rotors spinning. If you look at the reveal videos from that year, most teams highlighted the high crowd appeal value of shooting into the high boiler, but then when you watch actual game play, shooting was generally a secondary task. It is not clear to me whether the GDC over-estimated what teams would be capable of when it came to shooting that year or under-estimated how much time would need to be spent cycling gears to get all 4 rotors spinning.

The layout of the field can also have an effect on balance. Fields that offer more choke points along the typical cycling paths also tend to be the games where defense becomes a major factor in terms of winning or loosing an event. Wide open fields tend to favor offensive robots.

There is also balance from year to year. Having a mix of challenges between one season and the next forces teams to innovate rather than simply recycling the same design over and over. This is getting harder and harder to do as there are only so many ways you can design a game around the idea of shooting a round ball or picking and placing a non-ball object.

I think that the GDC did themselves a huge favor in 2016 when they gave themselves the ability to adjust the game balance as the season progressed. I’m surprised that they have not used this tactic more often. Every season the average score goes up and up from week to week as the season progresses. It is very predictable and if the game is designed to accommodate this, you could actually set a schedule in the manual at the beginning of the season about how the point values or RP achievement levels would change during the year. And if you gave yourself enough levers to pull you could change the reward for given tasks based on the actual difficulty of achieving them rather than simply your prediction of the relative difficulty when you designed the game.

7 Likes

GDC to this thread: :eyes:

2 Likes

But this wasn’t from a rule change - perhaps a place where a rule change would have been beneficial. Teams discovered that there was no rule preventing them from removing the gearboxes and running a super fast motor. I wonder how much a rule change requiring that the Tetrix gearboxes remain intact would have changed things in that game?

Here’s the presentation I gave on How to Design a Competitive Robotics Game (starts at around the 50:06 mark)

And the scoring analysis discussion starts at about 1:26:16. But basically when balancing a game, there are variety of factors to consider including:

  • Balancing design/build effort for the task with points
  • Balancing time spent on the task during the match with points
  • Balancing how much you want this task to happen with points
  • Balancing this task with all other tasks to ensure everything is scaled properly
10 Likes

Something that I think about off and on is how one would go about “beta-testing” a robotics game. I don’t think there’s a clear way to do this, but with the recent example we have (Infinite Recharge), at least in my eyes, the GDC was able to make minor tweaks that improved gameplay overall.

A lot of video games have open betas where players of all skill levels can play the game, report bugs, and give feedback to the devs before the final version is released.

In this sense, it can definitely be tricky for the GDC to balance a game when robot performance is difficult, if not impossible, to predict. There’s also probably not a clear solution to this either unfortunately.

I guess that’s where off-seasons fit in, as we can see many hosts making tweaks to improve gameplay like the Chezy Champs crew.

While explicitly allowing values to change it would complicate game analysis (which is a space that many teams already struggle) because now instead of “a low goal is worth X” where X is a discrete value you have a range of values with an unknown distribution. You’ve now added an entirely new set of assumptions that must be made.

The analysis of those assumptions drives design. If your analysis says that the best robot you can build is to shoot high and hang and you forego scoring low and then the GDC brings low goal points up what do you do?

The reason frequent adjustments are doable in the video game world is because, generally, the investment into a certain thing is minimal in real value. Sure someone may grind to unlock a weapon but most “nerf” patches are generally cautious except in the most egregious cases (The fact that the best reference I have is Modern Warfare 2 and the akimbo shotguns tells you how frequently I play competitive games…) But in modern games they have public test servers to ensure changes are somewhat balanced and many of the competitive games have release schedules and events are built around those.

But also, we have a 8 week comp season. Maybe you could do 2-3 tweaks in that span unless you want Week 8 comps to be entirely different games than Week 1 comps but given the number of teams that have large gaps without competing you also run into issues there.

Or the short version - changing game play inside an FRC season is rough. Once we define our core gameplay loops there is enough investment in there that changing them would need to be reserved for “this is ruining seasons” type gameplay. I can only thing of 2 times in the last 18 years I’d say we saw things like that:

469 - 2010: We all knew I was going to point it out. Many would say this bot broke the game and should have been nerfed. I played against it - it was a good bot with a mostly unique spin on the game. But it wasn’t 100% undefeated, it wasn’t impossible to beat. Just played the game well. And in the end - it didn’t win Champs.

254 - 2018: What no, it was just a great bot… no, it was, really. The issue wasn’t 254’s robot the issue was that the time based scoring coupled with the reinforcement mechanic of the tilting scoring platform meant that going into an early lead in auto and then just piling on largely made coming back difficult. They “exploited” a game mechanic that I think was poorly thought through (on FRC’s side). HQ’s hands were also largely tied - the mechanic was the core concept in their game… This was an example of where having solid simulation may have saved their bacon. It’s also super obvious in hindsight and less so looking at it initially.

3 Likes

I’m going to throw out a different take:
Points should not be balanced with effort.
Doing so just leaves struggling teams with nothing useful to do (2015) or forces teams into a “must do” ranking trap (2017, 2020).

Instead, every additional challenge from the “main cycle” should have diminishing returns. That way struggling teams can help provide the bulk of scoring. Meanwhile top teams will take every advantage, no matter how small. We really don’t have to worry about them being “fairly rewarded”. They’ll be fine!

9 Likes

The Steamworks scoring system is a pretty good example of what not to do. The unbalanced scoring values really harmed the quality of that game.

I think the diminishing returns concept can work, as long as things don’t diminish as drastically as they did in Steamworks after a team is able to do a climb and a few gears.

1 Like

I think the word balanced is a bit of a trap here - in most cases you’re not doing that you’re skewing the reward system to encourage the desired gameplay. It’s not really saying everything needs to be balanced (though I suppose in competitive shooters there is value in having some sense of rock/paper/scissors) but for FRC it’s less “all paths need to be viable” and more “this is the desired gameplay we want to see at our premiere events”

Maybe F1 is a better comparison tbh - their goal is to make the race work a certain way and they only have some points to change rules due to immense investments.

2 Likes

Not a post for/against COTS items, but I’m betting all the new stuff getting added really messes with handling this bullet point.

A lot of “climber in a box”(s) got sold this year. Does it turn into pay to win? Do you lose out on the engineering experience? Is it worth designing anymore? Not my questions, just some of the many questions and debates from COTS threads.

PS - please don’t try answering the questions. I just tossed them in there to point at some of the challenges associated with @Karthik’s bullet point

1 Like

While having COTS solutions that help simplify the design/build effort needed to perform a game task, it is not a slam dunk. Many teams spent a lot of time playing with the geometry of the climb, how to make the transfers from one rung to another and how to articulate the arms in order to achieve the desired climb mechanics.

Our team designed a telescoping tube climber for 2020 that we struggled to get/keep working. We learned a lot from that design. We had planned to buy a COTS climber this year, but that boat sailed without us. So we took our learnings from our 2020 climber and designed a climber design that uses flat plates (waterjet cut) with standoffs between them that allows us to use multiple bearings per stage to keep the stages from binding.

We had completed that design early and has the arms basically cut and assembled before we actually settled on our final climbing geometry (we had 3 or 4 configurations in play for most of the build season).

So having the arms completed was not enough to get us from the basic level of achievement on the climbing task (mid rung) to the higher levels of achievement (high rung and Trav rung). We needed to spend the extra time working out the details of how the arms would articulate to go from basic achievement to high achievement to high achievement in a short amount of game time (8-10 seconds to trav rung).

In terms of game balancing, the trav rung presented a challenge that rewarded teams that were able to solve that design puzzle. It further rewarded teams that put in the extra work to develop a solution that allowed room for the other robots to also fit on the trav rung.

So, the climbing task this year is a pretty good example of a task with multiple levels of difficulty involved that allowed teams to differentiate themselves and get rewarded for solving higher level problems (how do I leave room for the other robots on my alliance). It is tasks like this that make balancing the game easier. Once you have designed the task with the multiple levels of difficulty, then you need to figure out how to assign the rewards for achieveing the higher levels of difficulty. The hangar RP required at least 2 robots to be able to climb and at least one of them to achieve at least the high rung. As the season progressed and the level of game play improved, that task got easier to achieve. It would have been quite possible to change the point value needed to achieve the RP to require at least 1 robot on the trav rung or at least 2 robots on the trav rung in order to re-balance the game for the higher level of play.

2 Likes

You kinda hit my point exactly. And I agree, this year WAS a good example of multiple levels reward that I think does well with considering the COTS.

You could go with a straight out of the box, hit level 2 and be done (more or less). You could get two sets and design arm-over-arm. You could make your own and at a bare minimum hit the low bar.

Is that what the GDC was thinking about when they designed the game and were putting on the finishing touches and point values? I’d like to hope so.

Ahh 1887, a fine year. I can still hear the rage. I preferred the semtex, riot shield, javelin glitch